End To End QoS Network Design

End-to-End QoS Network Design
Tim Szigeti, CCIE No. 9794, and Christina Hattingh
Cisco Press
Cisco Press
800 East 96th Street
Indianapolis, IN 46240 USA
ii
End-to-End QoS Network Design

Tim Szigeti, CCIE No. 9794, Christina Hattingh
Copyright 2005 Cisco Systems, Inc.
Published by:
Cisco Press
800 East 96th Street
Indianapolis, IN 46240 USA
All rights reserved. No part of this book may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, recording, or by any information storage and retrieval system, without
written permission from the publisher, except for the inclusion of brief quotations in a review.
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
First Printing October 2004
Library of Congress Cataloging-in-Publication Number: 2003111984
ISBN: 1-58705-176-1
Trademark Acknowledgments
All terms mentioned in this book that are known to be trademarks or service marks have been appropriately capitalized. Cisco Press or Cisco Systems, Inc., cannot attest to the accuracy of this information. Use of a term in this book
should not be regarded as affecting the validity of any trademark or service mark.
Warning and Disclaimer

This book is designed to provide information about Quality-of-Service network design best-practice recommendations. Every effort has been made to make this book as complete and as accurate as possible, but no warranty or
tness is implied.
The information is provided on an as is basis. The authors, Cisco Press, and Cisco Systems, Inc., shall have neither
liability nor responsibility to any person or entity with respect to any loss or damages arising from the information
contained in this book or from the use of the discs or programs that may accompany it.
The opinions expressed in this book belong to the author and are not necessarily those of Cisco Systems, Inc.
Corporate and Government Sales

Cisco Press offers excellent discounts on this book when ordered in quantity for bulk purchases or special sales.
For more information please contact:
U.S. Corporate and Government Sales 1-800-382-3419
[email protected]
For sales outside the U.S. please contact: International Sales [email protected]
Feedback Information
At Cisco Press, our goal is to create in-depth technical books of the highest quality and value. Each book is crafted
with care and precision, undergoing rigorous development that involves the unique expertise of members from the
professional technical community.
Readers feedback is a natural continuation of this process. If you have any comments regarding how we could
improve the quality of this book or otherwise alter it to better suit your needs, you can contact us through e-mail
at [email protected]. Please make sure to include the book title and ISBN in your message.
We greatly appreciate your assistance.
iii
Publisher
Editor-in-Chief
Cisco Representative
Cisco Press Program Manager
Executive Editor
Acquisitions Editor
Production Manager
Development Editor
Copy Editor
Technical Editors
Team Coordinator
Cover Designer
Composition
Indexer
Proofreader
John Wait
John Kane
Anthony Wolfenden
Nannette M. Noble
Christopher Cleveland
Michelle Grandin
Patrick Kanouse
Howard A. Jones
Krista Hansing
Frank Knox
Anna To
Connie Varner
Tammi Barnett
Louisa Adair
Octal Publishing, Inc.
Eric Schroeder
Tonya Cupp
iv
About the Authors

Tim Szigeti, CCIE No. 9794, attended the University of British Columbia, where he majored in management
information systems. After graduating, Tim joined Cisco Systems and soon after began to specialize in Quality-ofService technologies, supporting technical marketing initiatives for the Cisco Class Data acquisition, which led to
the Cisco QoS Policy Manager (QPM) product. After supporting QPM through several generations and serving as
product manager for the Cisco Quality of Service Device Manager (QDM) product, Tim joined the Enterprise Solutions Engineering team and led large-scale testing initiatives of campus, WAN, and VPN QoS designs. Tim now
belongs to the newly formed Technology Solutions Engineering team within the Cisco Central Technical Marketing
organization. There, he continues to dene and drive strategic QoS solutions across Cisco technology groups and
business units while working with many Fortune 500 companiesboth enterprise and service providersproviding QoS design expertise.
Christina Hattingh is a member of the technical staff in the Multiservice Customer Edge Business Unit of Cisco
Systems. These products, including the Cisco 2600, 3600, and 3700 series access router platforms, were some of
the rst Cisco platforms to converge voice and data trafc onto an IP network by offering TDM voice interfaces,
WAN interfaces, and critical QoS features, while later integrating call control elements into the router-based platform itself. In this role, she trains Cisco sales staff and advises customers on voice network deployment and design.
About the Technical Editors

Frank Knox has more than 37 years of telecommunications experience. During his career at IBM, Frank held positions in eld service, eld support, service planning, and education; his nal position before retirement was curriculum manager for IBMs Network Education in North America. After leaving IBM, Frank held the position of
network engineering manager for GTE Directories, where he was responsible for the companys voice and data network design and support. Concurrent with his work at IBM and GTE, Frank taught as an adjunct professor for the
University of Dallas MBA program. For the past six years, Frank has worked for Skyline Computer as a senior
instructor and consultant; he is currently Skylines chief technical ofcer (CTO). Frank holds two CCIE certications (R&S and SNA/IP); he also has a masters degree in telecommunications from Pace University.
Anna To has worked with Cisco for more than three years as a software/deployment engineer on the ITD QoS
team. One of Annas key tasks is to promote QoS deployment and increase the understanding of QoS technology in
the eld. Anna works on the Modular QoS CLI (MQC) solution team to bring consistency in QoS conguration
across various Cisco platforms. In addition, Anna is involved with the AutoQoS project to simplify QoS deployment.
Connie Varner is a technical marketing engineer in the Cisco Enterprise Systems Engineering group. She has
extensive experience designing and testing large-scale networks based on customer requirements, in part based on
four years of experience with the Cisco Customer Proof of Concept Labs. Connie specializes in QoS designs that
meet the needs of converged data, voice and video networks, and designs that involve IPSec VPNs.
Dedications
Tim: This book is obviously dedicated to my wife; otherwise, of course, shed kill me. It amuses me to think that if
others are actually reading this, they probably think Im only jokingbut, alas, the Greek capacity for vengeance is
no laughing matter. I cancelled far too many dates, stayed in my ofce and labs far too many weekends, and stared
blankly into space (thinking about these designs) far too many times (while she was talking to me) to ever allow the
thought of not dedicating this work to her to even cross my tiny xeno-brain.
I know, I know, its not a work of literature or a collection of poetry: Its just a technical bookboring to tears for
any not interested in the subject (and probably just boring to yawns for the rest). But, for whatever its worth, Im
dedicating it to you, Lella. I love you with all my heart.
Christina: To Robert Verkroost and Ria and Willie Hattingh, who unfailingly support my various forays into the
publishing world.
vi
Acknowledgments
Off the top, Id like to thank my friend and co-worker Dave Barton, whoalthough he was extremely busy downing beers at Chicagos Navy Piergallantly managed to sic Brett Bartow onto me, which got the ball rolling on this
whole project. (Dave, did you make it back okay to the hotel that night?)
Many thanks to Todd Truitt, one of the top talents at Cisco, for inviting my collaboration on the original AVVID QoS
Design Guide, hiring me onto his design team, and recommending Christina as a co-author for this project. Do you
ever get tired of being right, Todd?
Thanks also to Neil Anderson, Joel King, Ted Hannock, and Steve Ochmanski for guidance and collaboration on
IPSec V3PN designs. Thanks for letting me leverage your excellent and thorough work so that I did not to have to
reinvent the wheel on these designs.
Thank you, Mike Herbert, for your brilliant ash of using QoS for DoS/worm mitigation via the Scavenger class.
Though you derailed and postponed many whitepapers and publications (including this one), you opened up a
whole new scope of application for QoS technologiesand were all better off for it.
Thank you, too, Alex Dolan, for building out multiple large-scale MPLS VPN testbeds for me and continually
tweaking them to suit my mood-of-the-day. I dont know where your patience or your good nature comes from, but
theyre most appreciated. Thanks, too, for nudging me back into playing ice hockey. Next time I break a leg or chip
a tooth, Ill think of you and grimace.
Muchos gracias, Arlindo Callejas, for being much more than my awesome lab administrator. You always went out
of your way for me and got me everything I ever neededinstantly. Sometimes Im afraid to ask where you
sourced the gear you did. (Im not sure whether those 10GE linecards fell off the back of a Cisco truck or what,
but they sure came in handy at just the right time.)
A round of applause is merited by the technical reviewers. Having done this before myself, I can genuinely appreciate the time, effort, and painstaking attention to detail that goes into this process. Frank, your comments were right
on and helped make this a better book. Anna, is there anything you dont know about Cisco QoS? Im very thankful
you took time out of your extremely busy schedule, developing code while helping anyone and everyone on planet
Earth (and some nearby systems) that are having QoS problems. And Connie, if you hadnt reviewed this work, I
would not have submitted it for publication. Youre simply the best technical reviewerand one of the sharpest
engineersIve ever had the pleasure of working with.
Thank you Howard Jones for your excellent editing and coordinating the complex content review and copy review
processes. And thank you, too, Patrick Kanouse for managing the production of this publication and allowing me to
make hundreds of last-minute edits in the galley-review phase (when edits are to be kept at a minimum). How you
put up with me I'll never know, but I truly appreciate your patience and desire to help make this book as correct and
as current as possible. Also thank you Chris Cleveland for your ne recommendations and guidance during the
course of production.
I need to extend thanks also to Debbie Morrison, who is, in my opinion, the best technical writerperiod. Debbie,
as Ive said over and over again, you polish my ugly little chunks of coal into beautiful diamonds. I love how I can
barely recognize my own work once youve done your magic. Ill truly miss working with you now that youve
gone on to bigger and better things. (Im so terried of the futurewhos going to make me look good now?)
Brett Bartow, what can I say? This would never have happened without you. Time and time again, it seemed to fall by
the wayside, but your persistence, perseverance, and patience kept it all going. Thank you. You didnt back off, and Im
glad for it. Your guidance has been uncanny, and your vision has paid off. Thanks also to your production team.
And lastly, thank you, Christina. You made it fun. Right when I read your rst draft of your rst chapter, I knew you
were the best person to embark on this project with (even though you write like an engineer!). Thank you for sacricing so many weekends on this (thank Robert for me too). I know this is only one of many publishing projects
youre pursuing; all I ask is that you save me an autograph before you move to Hawaii and start on your best-seller!
vii
Contents at a Glance
Introduction
xxii
Part I
Introduction to QoS
Chapter 1
Introduction to QoS
Chapter 2
QoS Design Overview
Part II
QoS Toolset
Chapter 3
Classication and Marking Tools
Chapter 4
Policing and Shaping Tools
Chapter 5
Congestion-Management Tools
Chapter 6
Congestion-Avoidance Tools
Chapter 7
Link-Specic Tools
Chapter 8
Bandwidth Reservation
Chapter 9
Call Admission Control (CAC)
Chapter 10
Catalyst QoS Tools
Chapter 11
WLAN QoS Tools
269
Part III
LAN QoS Design
287
Chapter 12
Campus QoS Design
Part IV
WAN QoS Design
Chapter 13
WAN Aggregator QoS Design
Chapter 14
Branch Router QoS Design
5
33
67
69
103
133
159
169
195
205
223
289
445
447
513
viii
Part V
VPN QoS Design
Chapter 15
MPLS VPN QoS Design
547
Chapter 16
IPSec VPN QoS Design
635
Appendix
QoS At-A-Glance Summaries
Index 713
545
701
ix
Table of Contents
Introduction
Part I
xxii
Introduction to QoS
Chapter 1 Introduction to QoS
A Brief Historical Perspective

QoS Evolution
User Network Expectations 9

End User 9
Information Technologies Management
Understanding QoS 10
End-to-End QoS 10
All Packets Are (Not) Equal 11
The Challenges of Converged Networks 12
QoS Models 14
IntServ Overview 15
DiffServ Overview 16
Introduction to the QoS Toolset
17
Simplifying QoS 19
Modular QoS Command-Line Interface 19
QoS Baseline 20
Default Behavior 21
Cross-Platform Feature Consistency 24
Automatic QoS 24
If I Have AutoQoS, Why Should I Be Reading This Book?
The Continuing Evolution of QoS 29
Summary
29
Further Reading 30
General 30
IntServ 30
DiffServ 31
AutoQoS 31
26
Chapter 2 QoS Design Overview
33
QoS Requirements of VoIP 33

Voice (Bearer Traffic) 33
Call-Signaling Traffic 38
QoS Requirements of Video
Interactive-Video 39
Streaming-Video 41
39
QoS Requirements of Data 42

Best-Effort Data 44
Bulk Data 44
Transactional Data/Interactive Data 45
Locally Defined Mission-Critical Data 45
DLSw+ Considerations 47
QoS Requirements of the Control Plane
IP Routing 48
Network-Management 49
48
Scavenger Class 49
DoS and Worm Mitigation Strategy Through Scavenger Class QoS 50
Principles of QoS Design 54
General QoS Design Principles 55
Classification and Marking Principles 57
Policing and Markdown Principles 57
Queuing and Dropping Principles 58
DoS and Worm Mitigation Principles 61
Deployment Principles 62
Summary
63
Further Reading
Part II
QoS Toolset
64
67
Chapter 3 Classification and Marking Tools
69
Classification Tools 70
Modular QoS Command-Line Interface Class Maps 71
Network-Based Application Recognition 73
xi
Marking Tools 77
Class-Based Marking 78
Class-Based Policing 78
Committed Access Rate 79
Policy-Based Routing 79
Voice Gateway Packet Marking 79
Layer 2 Marking Fields 81
Layer 3 Marking Fields 86
Translating Layer 2 and Layer 3 Packet Markings 90
Summary
98
Further Reading 99
General 99
DiffServ 99
L2 Protocol Tunneling 100
VPN 100
NBAR 100
MPLS 100
IPATM/Frame Relay Bundles 101
Level 2 to Level 3 Packet-Marking Translation
Chapter 4 Policing and Shaping Tools
101
103
Token Bucket Algorithms 105

Policers 107
Policers as Markers 107
Committed Access Rate 107
Class-Based Policing 109
Shapers 118
Shaping Algorithms 120
Shaping on ATM and Frame Relay Networks 121
Generic Traffic Shaping 126
Class-Based Shaping 126
Further Reading 128
DiffServ Policing Standards 128
Policing 129
ATM PVC Traffic Parameters 129
Frame Relay Traffic Shaping 129
Traffic Shaping 130
xii
Chapter 5 Congestion-Management Tools
133
Understanding Scheduling and Queuing
134
Legacy Layer 3 Queuing Mechanisms 136

Priority Queuing 137
Custom Queuing 137
Weighted Fair Queuing 137
IP RTP Priority Queuing 139
Currently Recommended Layer 3 Queuing Mechanisms 139
Class-Based Weighted Fair Queuing 139
Low-Latency Queuing 140
Layer 2 Queuing Tools 150
Frame Relay Dual-FIFO 150
PVC Interface Priority Queuing
Tx-ring
152
PAK_priority
Summary
150
153
154
Further Reading 154

Layer 3 Queuing 154
Layer 2 Queuing 155
Tx-ring 156
PAK_priority 156
Chapter 6 Congestion-Avoidance Tools
Random Early Detection
159
160
Weighted Random Early Detection
161
DSCP-Based Weighted Random Early Detection

Explicit Congestion Notification
Summary
163
166
Further Reading 166

DiffServ Standards Relating to WRED 166
Cisco IOS WRED Documentation 166
162
xiii
Chapter 7 Link-Specific Tools
169
Header-Compression Techniques 170

Related Standards 171
TCP Header Compression 171
RTP Header Compression 171
Compression Formats 173
Layer 2 Encapsulation Protocol Support 174
Summary of cRTP Formats and Protocol Encapsulations 177
Class-Based Header Compression 178
Advanced Topics on cRTP 180
Link Fragmentation and Interleaving 181
Fragment Sizes 183
Multilink PPP LFI 183
Frame-Relay Fragmentation 185
LFI for Frame Relay/ATM Service Interworking
IPSec Prefragmentation 190
Summary
190
Further Reading 191

General 191
IETF Standards 191
Frame Relay Forum Standards 191
Header Compression 192
Link Fragmentation and Interleaving 192
Chapter 8 Bandwidth Reservation
195
RSVP Overview 196

RSVP Service Types 197
Admission Control 197
RSVP and LLQ 198
MPLS Traffic Engineering
Scalability
199
200
RSVP-DiffServ Integration
200
Endpoints and Proxies 201

Summary
201
Further Reading 201

Standards 201
Cisco IOS Documentation
202
188
xiv
Chapter 9 Call Admission Control (CAC)
205
CAC Overview 205

CAC Defined
206
CAC Tool Categories 207

Local CAC Tools 208
Measurement-Based CAC Tools 208
Resource-Based CAC Tools 208
CallManager Locations CAC
Gatekeeper CAC
209
211
RSVP 212
Example of VoIP CAC Through RSVP 215
Summary
218
Further Reading 218

General 218
Chapter 10 Catalyst QoS Tools
218
223
Generic Catalyst QoS Models 224

Classification, Marking, and Mapping
Policing and Markdown 227
Queuing and Dropping 228
224
Catalyst 2950 231

Catalyst 2950 Classification, Marking, and Mapping
Catalyst 2950 Policing and Markdown 234
Catalyst 2950 Queuing 235
Catalyst 3550 235
Catalyst 3550 Queuing and Dropping 240
232
237
Catalyst 2970, 3650, and 3750 242

Catalyst 2970/3560/3750 Classification, Marking, and Mapping
Catalyst 2970/3560/3750 Policing and Markdown 244
Catalyst 2970/3560/3750 Queuing and Dropping 244
Catalyst 4500 247
248
243
xv
Catalyst 6500 252

Summary
254
263
Further Reading
266
Chapter 11 WLAN QoS Tools
269
QoS for Wireless LANs Versus QoS on Wired LANs 270

Upstream Versus Downstream QoS 271
IEEE 802.11 DCF 272
Interframe Spaces 272
Random Backoffs/Contention Windows 273
IEEE 802.11e EDCF 275
QoS Basic Service Set Information Element
IEEE 802.1D Classes of Service
278
279
QoS Operation on Cisco APs 280

Configuring QoS on Cisco APs 281
Summary
284
Further Reading
Part III
LAN QoS Design
285
287
Chapter 12 Campus QoS Design
289
DoS/Worm-Mitigation Strategies 292

Scavenger-Class QoS Operation 294
Call-Signaling TCP/UDP Ports in Use
295
Access-Edge Trust Models 302

Trusted Endpoint Models 302
Untrusted Endpoint Models 304
Conditionally Trusted Endpoint(s) Models 307
Catalyst 2950 QoS Considerations and Design 314
Catalyst 2950: Trusted Endpoint Model 314
Catalyst 2950: Untrusted PC with SoftPhone Model 315
Catalyst 2950: Untrusted Server Model 315
Catalyst 2950: Conditionally Trusted IP Phone + PC: Basic Model
319
xvi
Catalyst 2950: Conditionally Trusted IP Phone + PC: Advanced Model

Catalyst 2950: Queuing 322
322

Catalyst 3550: Conditionally Trusted IP Phone + PC: Basic Model 331
Catalyst 3550: Conditionally Trusted IP Phone + PC: Advanced Model 333
Catalyst 3550: Queuing and Dropping 336
Catalyst 2970/3560/3750 QoS Considerations and Design 342
Catalyst 2970/3560/3750: Trusted Endpoint Model 343
Catalyst 2970/3560/3750: Untrusted PC with SoftPhone Model 344
Catalyst 2970/3560/3750: Untrusted Server Model 345
Catalyst 2970/3560/3750: Conditionally Trusted IP Phone + PC: Basic
Model 346
Catalyst 2970/3560/3750: Conditionally Trusted IP Phone + PC: Advanced
Model 348
Catalyst 2970/3560/3750: Queuing and Dropping 351
Catalyst 4500-SupII+/III/IV/V QoS Considerations and Design 357
Catalyst 4500: Queuing 366
Catalyst 6500: CatOS Defaults and Recommendations 375
Catalyst 6500: Queuing and Dropping 391
Catalyst 6500: PFC3 Distribution-Layer (Cisco IOS) Per-User Microflow
Policing 419
WAN Aggregator/Branch Router Handoff Considerations 420
Case Study: Campus QoS Design
Summary
440
Further Reading
441
422
xvii
Part IV
WAN QoS Design
445
Chapter 13 WAN Aggregator QoS Design
447
Where Is QoS Needed over the WAN?
447
WAN Edge QoS Design Considerations 448

Software QoS 448
Bandwidth Provisioning for Best-Effort Traffic 449
Bandwidth Provisioning for Real-Time Traffic 449
Serialization 450
IP RTP Header Compression 451
Tx-ring Tuning 451
PAK_priority 452
Link Speeds 452
Distributed Platform QoS and Consistent QoS Behavior
WAN Edge Classification and Provisioning Models 453
Slow/Medium Link-Speed QoS Class Models 454
High Link Speed QoS Class Models 459
WAN Edge Link-Specific QoS Design 467
Leased Lines 467
Frame Relay 478
ATM 488
ATM-to-Frame Relay Service Interworking
ISDN 501
497
Case Study: WAN Aggregation Router QoS Design

Summary
505
507
Further Reading
508
Chapter 14 Branch Router QoS Design
513
Branch WAN Edge QoS Design 514

Unidirectional Applications 514
Branch Router LAN Edge QoS Design 517
DSCP-to-CoS Remapping 518
Branch-to-Campus Classification and Marking 519
NBAR Known-Worm Classification and Policing 526
Case Study: Branch Router QoS Design
Summary
541
Further Reading
541
535
453
xviii
Part V
VPN QoS Design
545
Chapter 15 MPLS VPN QoS Design
547
Where Is QoS Needed over an MPLS VPN?
548
Customer Edge QoS Design Considerations 550

Layer 2 Access (Link-Specific) QoS Design 550
Service-Provider Service-Level Agreements 551
Enterprise-to-Service Provider Mapping Models 552
Provider-Edge QoS Considerations 563
Service Provider-to-Enterprise Models 563
MPLS DiffServ Tunneling Modes 566
Core QoS Considerations 582
Aggregate Bandwidth Overprovisioning
DiffServ in the Backbone 583
MPLS Traffic Engineering 587
583
Case Study: MPLS VPN QoS Design (CE/PE/P Routers)

Summary
616
632
Further Reading
632
Chapter 16 IPSec VPN QoS Design
635
Site-to-Site V3PN QoS Considerations 637

IPSec VPN Modes of Operation 637
Packet Overhead Increases 640
cRTP and IPSec Incompatibility 643
Prefragmentation 644
Bandwidth Provisioning 645
Logical Topologies 646
Delay Budget Increases 647
ToS Byte Preservation 648
QoS Pre-Classify 649
Pre-Encryption Queuing 651
Anti-Replay Implications 654
Control Plane Provisioning 657
Site-to-Site V3PN QoS Designs 658
Headend VPN Edge QoS Options for Site-to-Site V3PNs 665
xix
Teleworker V3PN QoS Considerations 666

Teleworker Deployment Models 667
Broadband-Access Technologies 671
Bandwidth Provisioning 674
Asymmetric Links and Unidirectional QoS 677
Broadband Serialization Mitigation Through TCP Maximum Segment Size
Tuning 678
Split Tunneling 679
Teleworker V3PN QoS Designs 682
Case Study: IPSec VPN QoS Design
Summary
696
Further Reading
697
Appendix QoS At-A-Glance Summaries

Index 713
701
686
xx
Icons Used in This Book

Communication
Server
PC
PC with
Software
Terminal
File
Server
Sun
Workstation
Macintosh
Access
Server
Cisco Works
Workstation
Modem
Token
Ring
Token Ring
Printer
Laptop
Web
Server
IBM
Mainframe
Front End
Processor
Cluster
Controller
FDDI
Gateway
Router
Catalyst
Switch
Network Cloud
Bridge
Multilayer
Switch
Line: Ethernet
Hub
ATM
Switch
Line: Serial
DSU/CSU
DSU/CSU
FDDI
ISDN/Frame Relay
Switch
Line: Switched Serial
xxi
Command Syntax Conventions

The conventions used to present command syntax in this book are the same conventions used in the
Cisco IOS Command Reference. The Command Reference describes these conventions as follows:
Boldface indicates commands and keywords that are entered literally as shown. In actual
conguration examples and output (not general command syntax), boldface indicates commands
that are input manually by the user (such as a show command).
Italics indicates arguments for which you supply actual values.
Vertical bars (|) separate alternative, mutually exclusive elements.
Square brackets [ ] indicate optional elements.
Braces { } indicate a required choice.
Braces within brackets [{ }] indicate a required choice within an optional element.
xxii
Introduction
QoS is a maturing technology, one that many networking professionals, to a greater or lesser extent, are
already familiar with. This is both a blessing and a curse. It is a blessing because more administrators
are enabling QoS on their networks, which allows for the convergence of voice, video, and data onto a
single IP network, among other business advantages. It is a curse because almost every individual with
whom Ive ever discussed QoS designs has a slightly different opinion on how QoS should be enabled.
The result often has led to confusing babble from the customers perspective, especially for customers
seeking QoS design guidance for non-VoIP applications. For example, a customer might ask the local
Cisco Systems engineer how best to enable QoS for networks and receive one answer. Later, the customer might attend an Executive Brieng session in San Jose and receive a different answer (even
receiving multiple different answers within the same day from different presenters). Later, while attending
a Networkers conference, the customer might be told something else entirely. Finally, when the customer
gets home and picks up a Cisco Press book, he or she might get still another story. Confused and frustrated, many customers decide to enable minimal QoS, if any, despite the touted benets that they were
sold on. Therefore, in my opinion, presenting such inconsistent recommendations is a major disservice
to our customers and a considerable barrier to the widespread deployment of QoS.
The Cisco Technology Baseline committees were created to remedy the situation and help unify various
technologies across Cisco products and platforms. To this end, a series of Technology Baselines were
developed internally by our leading experts (many of whom likewise developed the related IETF RFCs
and other standards) to which all Cisco products and features must conform. Additionally, these documents provide uniform, strategic recommendations (that can be shared with customers) to help ensure
that QoS recommendations are unied and consistent, for both enterprises and service providers. Specic to QoS, the QoS Baseline strictly denes the Cisco strategic direction in QoS technologies from
now into the foreseeable future.
Thus, a unique feature of this book is that it is the rst Cisco Press publication to present design recommendations that are compliant with the QoS Baseline.
Another huge advantage of this publication is that it is one of the rst documents to present a detailed,
cohesive strategy that shows how QoS can extend beyond its traditional role (of prioritizing important
applications) and be used to provide deferential services to DoS/worm-generated trafc, thus mitigating
and containing the collateral damage caused by such attacks. This is a fresh perspective and context for
a technology that many considered baked and done. Yet in such a role, the critical interdependency of
Quality of Service, High-Availability, and Security technologies becomes manifest and holistically promotes the Self-Defending Networks business objective.
However, having a strategic direction and tactical approaches for QoS designs is only half the solution.
An important motto that I like to emphasize is: In theory, theory and practice are the same. Its one
thing to make a design recommendation based on an assumption that something should work. Its
something completely different to make a design recommendation that has been veried in large-scale,
complex lab scenarios, such as provided by one of the largest Cisco labs: the Enterprise Solutions
Engineering testbeds in Research Triangle Park, North Carolina.
xxiii
Notwithstanding, it should be noted that designs presented in this book are not infallible. While all due
diligence has been done to present working, tested congurationsincluding a rigorous technical
reviewing process by some of the sharpest Cisco QoS engineershardware/software/platform-specic
issues that didnt surface during our tests may nonetheless exist, as may issues introduced in newer
releases of hardware/software dating from our time of testing.
Furthermore, the recommendations presented in this book are not to be taken as commandments or
dictates (Thou shalt congure this or that), but are simply best-practice design recommendations that
are the result of extensive lab testing and customer deployments. They should be viewed as templates
that can be modied and tweaked to customer-specic requirements. Following the 80/20 Pareto Rule,
these design recommendations should be viewed as 80 percent of the solution, to which the remaining
20 percent is up to each customer to complete and tailor to their individual needs and constraints.
Heres an analogy of how to view these design recommendations: Given a business objective (for example, to hammer a nail into a wall), you will have certain tools at your disposaltools that may or may
not be optimally suited to the task (lets say, a hammer and a banana). Our lab testing presents the optimal
tool to use for the given objective (normally, a hammer tests better than a banana, but you never know
Ive seen some pretty funky frozen bananas that might do the trick). Its still up to the customer to pick
the tool that best suits their objectives, situations, and comfort levels. These recommendations are not
mandates; they are simply suggestions based on extensive lab testing and customer deployments.
Who Should Read This Book?

Some might ask, Why should I read this book? Especially when I have AutoQoS?
Certainly, AutoQoS-VoIP is an excellent tool for customers whose objective is enabling QoS for VoIP
(only) on their campus and WAN infrastructures, and AutoQoS-Enterprise is a ne tool for enabling
basic WAN-edge QoS for voice, video, and multiple classes of data. For customers who have basic QoS
needs and dont have the time or desire to learn or do more with QoS, AutoQoS is denitely the way
to go.
However, its important to remember where AutoQoS came from. AutoQoS tools are the result of QoS
design guides that Cisco Technical Marketing Engineers (including myself) put together based on largescale lab testing. AutoQoS-VoIP is the product of our rst AVVID QoS Design Guide, one of the most
popular and most downloaded technical whitepapers ever produced within Cisco. AutoQoS-Enterprise
is the result of the QoS Baseline coupled with our second-generation QoS Design Guide. This book represents our third-generation QoS Design Guide. And it is the goal of the authors to drive these designs
(including DoS/worm-mitigation strategies) into future releases of AutoQoS. So, basically, what you are
reading is the proposed blueprint for the next version of AutoQoS.
When it comes to any given technology, there are really only two types of people: those who are interested in the technology and seek a thorough understanding of the relation of the parts to the whole, and
those who just want to turn it on and walk away. The former are the ones who will condently unleash
the true power of the technology and push it to its limits; the latter are the ones who are usually hesitant,
timid, and conservative in their use of the technology, typically accompanied with mediocre results.
For example, there are those who enjoy looking under the hood of a Ferrari and want to know all the
details about how the engine generates its beautiful purring and power, and there are others who want
xxiv
only to turn it on, drive away, and look sexy. The former group will drive more condently, boldly
unleashing the engines tremendous power and, thus, pushing the car to its limits.
This book is intended for the former type of QoS networking professionalthose looking for a thorough
understanding of what makes them move so fast, sound so good, and look so sexy as they condently
harness their technology.
Goals and Methods

The main goal of this book is to present templates that address 80 percent or more of a customers
requirement of QoS in a particular context and architecture (LAN, WAN, VPN). Additionally, the
rationales and considerations behind the recommendations are explained in detail so that as tweaking is
required, network administrators are well informed of the trade-offs involved.
A key approach that weve used throughout this conguration-rich book is to incorporate inline explanations of congurations. In this way, the QoS-relevant commands are highlighted and detailed line-byline to illustrate the function of each element and how these parts make up the whole solution.
To complement these line-by-line design recommendations, related verication commands are detailed.
These verication commands are presented in context with the design examples, and specic details of
what to look for in the resulting output are highlighted. These verication examples are, therefore, signicantly richer in relevance than most such examples presented in Cisco documentation, and they
allow network administrators to conrm quickly whether the recommended designs have been deployed
correctly.
Finally, each design chapter has a case-study example at the end that ties together many of the design
elements presented in the chapter and presents a bigger-picture detailed example for the infrastructure
architecture being discussed (LAN/WAN/VPN). These examples are indicative of what can be expected
in production environments. Often these case-study examples span several devices and, thus, highlight
critical interrelationships.
How This Book Is Organized

This book is divided into three main parts: an introduction and overview section, a QoS toolset review
section, and (the heart of the book) a QoS design section.
Chapter 1, Introduction to QoS, is an introduction and brief history of the development of QoS
technologies, showing where these came from and the direction theyre headed in.
Chapter 2, QoS Design Overview, is an overview of QoS design. It begins by detailing the
service-level requirements of voice, video, and data applications, and it presents the Scavengerclass DoS/worm-mitigation strategy and high-level QoS best practices that will be detailed in the
design chapters to follow.
To set proper context for the design chapters, various QoS tools are reviewed. This review is not indented
to serve as feature documentation, but it supplements Cisco documentation to highlight various interdependancies or caveats for these tools that at times impact the recommended QoS designs that follow.
The QoS toolset review section, Chapters 3 through 11, covers the following topics:
xxv
Chapter 3, Classication and Marking ToolsThis chapter reviews Layer 2 marking

mechanisms (such as 802.1Q/p, Frame Relay Discard Eligibility, ATM Cell Loss Priority, and
MPLS Experimental Values) and Layer 3 marking mechanisms (such as IP Precedence and
Differentiated Services Code Points).
Chapter 4, Policing and Shaping ToolsThis chapter reviews the token bucket algorithm,
which is the basis for most policers and shapers. Both two-rate and three-rate policers are covered
as are ATM and Frame Relay trafc shaping.
Chapter 5, Congestion-Management ToolsThis chapter reviews the evolution of queuing

mechanisms and focuses on Low-Latency Queuing and Class-Based Weighted Fair Queuing. This
chapter highlights the interoperation and interdependencies of these mechanisms with other QoS
mechanisms, such as link-fragmentation and shaping tools.
Chapter 6, Congestion-Avoidance ToolsThis chapter reviews the Weighted Random Early

Detection mechanism and shows how this can be used to provide Differentiated Services within an
(RFC 2597) Assured Forwarding trafc class. This chapter also shows how this mechanism can be
used to set (RFC 3168) IP Explicit Congestion Notication bits.
Chapter 7, Link-Specic ToolsThis chapter reviews header-compression techniques (such as

TCP and RTP header compression) and link-fragmentation and interleaving techniques (such as
Multilink PPP Link Fragmentation and Interleaving [MLP LFI] and Frame Relay fragmentation
[FRF.12]).
Chapter 8, Bandwidth ReservationThis chapter reviews the Resource Reservation Protocol

(RSVP) and shows how it can be applied to admission control and MPLS Trafc Engineering.
Chapter 9, Call Admission Control (CAC)This chapter reviews local, resource-based, and
measurement-based call admission control (CAC) mechanisms, including the use of RSVP for
CAC. The tools reviewed in previous chapters can protect voice from data, but only CAC tools can
protect voice from voice.
Chapter 10, Catalyst QoS ToolsThis chapter reviews the main classication, marking,
mapping, policing, and queuing tools available on the current Cisco Catalyst platforms (including
the Catalyst 2950, 2970, 3550, 3560, 3570, 4500-Supervisors II+ to V, and Catalyst 6500
Supervisor 2 and Supervisor 720).
Chapter 11, WLAN QoS ToolsThis chapter reviews QoS mechanisms available for wireless
access points, including the 802.11e Enhanced Distributed Coordination Function (EDCF) and the
QoS Basic Service Set (QBSS).
When the QoS toolset is reviewed, the context is set for the detailed design recommendations that
follow. The next chapterswhich comprise the heart of this bookcover the QoS design recommendations for protecting voice, video, and multiple classes of data while mitigating DoS/worm attacks for the
following network infrastructure architectures:
Chapter 12, Campus QoS DesignThis design chapter details access, distribution, and core
layer considerations and designs for Cisco Catalyst 2950, 2970, 3550, 3560, 3570, 4500-Supervisors
III-V, and Catalyst 6500 Supervisor 2 and Supervisor 720 series switches. Five separate access-edge
xxvi
models are presented, along with detailed queuing/dropping recommendations on a per-platform

basis. Platform-unique features, such as the Catalyst 3550 per-Port/per-VLAN policing feature, the
Catalyst 6500 PFC2 Dual-Rate Policing feature, and the PFC3 Per-User Microow Policing
feature, are highlighted in context.
Chapter 13, WAN Aggregator QoS DesignThis design chapter details considerations and
designs for low-speed ( 768 kbps), medium-speed (> 768 kbps and T1/E1), and high-speed
(> T1/E1) private WAN topologies, such as leased lines, Frame Relay, ATM, ATM-to-Frame Relay
service interworking, and ISDN.
Chapter 14, Branch Router QoS DesignThis design chapter details branch-specic
considerations and designs, such as unidirectional applications, and branch-to-campus trafc
classication through access lists and Network-Based Application Recognition (NBAR). Branchspecic designs include Cisco SAFE recommendations for using NBAR for known worm
identication and policing.
Chapter 15, MPLS VPN QoS DesignThis design chapter details considerations and designs
for both enterprises (that are mapping into MPLS VPN service-provider [edge] classes of service)
and service providers (that are provisioning edge and core classes of service). Service provider
designs also include details on how to provision MPLS DiffServ Tunneling Modes (Uniform, ShortPipe, and Pipe) and an introduction to MPLS Trafc Engineering (demonstrating per-customer trafc
engineering and per-customer/per-application trafc engineering through MPLS DiffServ Trafc
Engineering).
Chapter 16, IPSec VPN QoS DesignThis design chapter details the considerations and
designs for deploying site-to-site IPSec VPNs and for teleworker IPSec VPNs (which traverse
broadband media, such as cable and DSL).
Appendix, At-a-Glance QoS SummariesSingle-page summaries of key QoS concepts

presented throughout this the book for ready-reference, including
QoS Tools
The Cisco QoS Baseline
QoS Best Practices
Scavenger-Class QoS Design
Campus QoS Design
WAN QoS Design
Branch QoS Design
MPLS VPN QoS Design (for Enterprise Subscribers)
MPLS VPN QoS Design (for Service-Providers)
This page intentionally left blank
PA R T
Introduction to QoS
Part I of this book provides a brief background of the evolution of QoS technologies and overviews
various currently available QoS features and tools. The QoS requirements of voice, video, and multiple
classes of data applications are presented, along with an overview of the nature and effects of various types
of DoS and worm attacks. QoS design principles are introduced to show how QoS mechanisms can be
strategically deployed to address application requirements while mitigating such attacks.
The chapters in this part of the book are as follows:
Chapter 1
Introduction to QoS
Chapter 2
QoS Design Overview
This chapter provides a brief history of both voice and data network evolution, illustrating
the background of the concept of Quality of Service (QoS). It introduces the fundamental
QoS concepts of IntServ and DiffServ to set the stage for the later discussion of individual
QoS mechanisms. The following topics are introduced and briey discussed:
IntServ and DiffServ

QoS tools categories
Modular QoS CLI
QoS Baseline
QoS classes
Automatic QoS
CHAPTER
Introduction to QoS
A fair amount has been written in the industry about quality of service (QoS), but it seldom
is explained why QoS has become a topic of concern in modern networks when it was
something relatively unheard of only a few years ago. It is instructive to review briey
a small amount of networking history that puts QoS technology in perspective.
A Brief Historical Perspective

A century ago, the public switched telephone network (PSTN) started building out a worldwide, circuit-switched network. This network consisted of xed-bandwidth, dedicated circuits and ideally was suited to carrying real-time trafc, such as voice. Some ve decades
later, networking experts from military and educational environments introduced packetswitched networks to circumvent any single points of failure, common in circuit-switched
networks. Packet switching chops the information ow into small chunks of data, which
can be routed over independent paths to the same destination, analogous to the operation of
the postal system.
The resiliency of packet-switched networks caused a shift toward connectionless communication protocols that can handle packets that might arrive out of order. However, for many
data applications, this was not only complicated to design around, but it also was insufcient in meeting application needs. Thus, connection-oriented protocols such as X.25 and
Systems Network Architecture (SNA), and later Frame Relay and Asynchronous Transfer
Mode (ATM), were developed. In these protocols, a circuit (permanent or switched virtual
circuit, PVC or SVC) is dened over the underlying connectionless packet-switched network to handle a session of communication between two devices or endpoints.
In theory, real-time communications such as voice can use virtual circuits regardless of the
underlying networking technology. Several universities conducted many experiments to
this effect in the 1970s. However, voice over virtual circuits did not become a commercial
or practical reality until the routers and switches used in packet-switched networks gained
the CPU and memory power required to drive packet streams at real-time speeds at costeffective prices.
When processing power became available at affordable cost points, other issues with
carrying real-time communications over a packet-switched network manifested themselves.
For example, when packets are delayed or dropped en route because of buffer overows or
Chapter 1: Introduction to QoS
other momentary failures, the intelligent protocols in the seven-layer International

Organization for Standardization (ISO) model recover the session through the use of
error-detection and correction capabilities, such as timeouts and retransmissions. Although
these recovery methods work well for data applications, they fall far short of satisfying the
needs of real-time information streams.
ATM was the rst general data-networking technology to include a class of service concept
at the lower layers of communications transport protocolsthat is, offering different
treatments for different types of trafc (protocols such as SNA already had this concept at
higher layers in the stack). ATM minimizes latency by dening xed-length cells, which
can be switched in hardware. ATM also uses the familiar concepts of PVCs and SVCs to
eliminate routing delays by doing an initial connection setup, after which all other packets
that belong to the stream can be forwarded to the destination without additional routing.
It has been saidonly partly facetiouslythat packet switching has been a 30-year failed
experiment. The attributes that the ATM architecture strived for are none other than those
of the original circuit-switched PSTN: low latency, xed circuit-based routing, predictable
service levels, and information-order preservation. Then why not just use the PSTN?
Options for transmitting data over the voice network met with equally limited success.
The PSTN simply is not optimized for data networks: The equipment is expensive, the
architecture is rigid, the bandwidth allocations are wasteful, and the infrastructure as a
whole is ill suited to applications in which sessions are short, variable, multipoint, or
connectionless.
In an effort to nd a more effective solution to support a combination of voice and data,
integrated services offerings were introduced. The term integrated services (as in Integrated
Services Digital Network [ISDN] or the IETF Integrated Services [IntServ] model) refers
to the mixing of different types of trafc, such as voice, video, and data, over a single
packet-switched network. ISDN, which is considered by some to be the rst foray into an
architecture for converged networks, denes how data protocols can be carried over a
circuit-switched infrastructure that natively supports voice. The PSTN evolved slightly
with the introduction of ISDN. However, with the exception of a handful of countries in
Europe, ISDN never really took off. This was primarily because of nontechnical reasons,
such as the pricing strategies that the providers adopted.
In the late 1990s, IP won out as the technology of choice for converged networks because
of its ease of use, ubiquity, and advances in handling real-time trafc.
The key enablers for IP networks to converge successfully voice, video, and data over
packet-switched infrastructures were QoS technologies. QoS allows for the differentiated
treatment of data trafc versus voice and other delay-sensitive trafc. In other words, it
allows the network to include a system of managed unfairness. As QoS technologies
evolved, increasing numbers of enterprises saw the value of a single-network infrastructure
and began planning toward deploying converged networks.
QoS Evolution
QoS Evolution
IP networks of the mid-1990s were invariably best-effort networks, and the Internet, as a
whole, remains so today. Privately owned enterprise and service provider networks, though,
have been widely transformed from best-effort models to more complex differentiated
services models, meaning that the network gives different applications differing levels
of service.
Figure 1-1 shows the broad steps in the evolution of QoS concepts since the early 1990s.
QoS Evolution
QoS as a
Security Tool
Increased Sophistication
Figure 1-1
QoS Intelligence
and Automation
MPLS Traffic Engineering
and VPN QoS
Differentiated Services
Model
Integrated Services
Model
Best Effort
IP Model
1994
1996
1998
2000
2002
2004
The rst attempt to standardize QoS came in the mid-1990s, when the Internet Engineering
Task Force (IETF) published the Integrated Services Request For Comments (IntServ
RFCsstarting with RFC 1633 in June 1994). These RFCs centered on a signaling protocol called the Resource Reservation Protocol (RSVP). RSVP signals bandwidth and latency
requirements for each discrete session to each node along a path (logical circuit) that
packets would take from the sending endpoint to the receiving endpoint. Initially, RSVP
required every node to heed its reservations, which was highly impractical over the Internet,
on which servers, switches, and routers of every description, vintage, and vendor coexist.
To address this challenge, another set of standardsthe DiffServ modelemerged as a
second attempt at standardizing QoS. The DiffServ model describes various behaviors to
be adopted by each compliant node. The nodes could use whatever features (proprietary or
otherwise) were available, as chosen by the vendor, to conform. Packet markings, such as
IP Precedence (IPP) and its successor, Differentiated Services Code Points (DSCPs), were
dened along with specic per-hop behaviors (PHBs) for key trafc types.
As the IntServ and DiffServ models have evolved, the general popularity of one method
versus the other has swung back and forth (as shown in Figure 1-2), and their coexistence
has become an ever-increasing struggle, with committed advocates on both sides. Today the
debates over the advantages of each continue without a clear, industry-wide agreed-upon
resolution. The realization has begun that neither method offers a complete solution and
that elements of both should be combined to provide the most general method applicable
to the widest range of trafc and application types.
Figure 1-2
IntServ and DiffServ
Integrated Services
Per Flow State
RSVP as the
Signaling Protocol
Differentiated Services
No State, No Protocol
Per Aggregate Behavior QoS
Less Scalability
More Scalability
A short denition of IntServ and DiffServ follows:
IntServ uses a ow-based concept coupled with a signaling protocol along the packet
path. The signaling protocol guarantees that adequate resources are available (at each
hop) for the ow before admitting the ow onto the network. In initial deployments,
the IntServ model suffered from scalability issues because of the many ows that
needed management on network backbones.
DiffServ uses packet markings to classify and treat each packet independently.
Although this scales well (which is probably why enterprises and service providers
deploy it more frequently), it offers no specic bandwidth guarantees to packets that
belong to a ow and, therefore, fails to provide admission control to new ows.
With no clear advantage to either model, QoS mechanisms continue to use a mix of IntServ
and DiffServ technologies to offer the breadth of services required on networks. IntServ and
DiffServ are discussed further in the section QoS Models later in this chapter.
User Network Expectations
In the late 1990s, QoS techniques became more sophisticated and were adapted to advanced
networking technologies, such as Multiprotocol Label Switching (MPLS) and Virtual
Private Networks (VPNs).
The most recent trend in QoS is simplication and automation, with the goal of simply and
efciently provisioning intelligent QoS on IP networks. When viewed as individual
features, QoS technologies offer a myriad of nerd knobs that can be turned. In the hands
of capable administrators, these can be used to build very sophisticated networks. However,
they also can result in very complex congurations. Many administrators do not have the
time or the desire to delve into QoS technologies to this expert level and instead would
prefer to dene high-level policies and have the network simply do the right things
to implement them.
User Network Expectations

The perception of how the network behaves, or how well it behaves, depends on how you
look at it. End users, or consumers, of the network might have a very different view of
network quality of service than the staff managing the network.
End User
An end users perception of QoS is typically subjective and relative, and not easily
measurable.
End users perceive the network quality through their end device and have certain expectations of appropriate service levels. For example, typical users expect a voice call on a
standard phone to be of toll quality. Yet they do not expect that same quality level from a
cell phone or a voice call spoken into a microphone on a personal computer. This is the general expectation, regardless of the fact that all three calls might be carried by the same network in the same manner.
The end user has no concept of and typically very little interest in the capabilities of the
networks in betweenunless, of course, the quality of the network response is lacking.
Therefore, end users perception of the quality of the service that they receive is based on
previously observed behavior on similar devices (examples include a PSTN phone call, a
PBX phone call, a bank teller application, and the refresh of a web page display) and the
cost of the service (which, in an enterprise network, the general end user perceives as $0).
Information Technologies Management

IT management perceives the network quality through management statistics such as
throughput, usage, percentage of loss, and user complaints. Expectations and quality
problems from an IT perspective tend to be more absolute and measurable. (For example,
10
the one-way latency of more than 150 ms is unacceptable for a voice call, or a transaction
refresh time for a bank teller application must be fewer than 5 seconds.)
Service providers formalize these expectations within service-level agreements (SLAs),
which clearly state the acceptable bounds of network performance. Corporate enterprise
networks typically do not have such formal SLAs, but they nevertheless manage their
networks on some form of measurement and monitoring. Some enterprise networks indeed
might use SLAs of various levels of formality between the IT department and the customer
departments that they serve.
User complaints might result even though the network met the SLA (formal or otherwise).
For example, the packet loss statistics over a certain period of time might be within bounds,
but the user whose le transfer session timed out or whose print job was lost during a period
of momentary congestion would most likely perceive it as a network problem.
Understanding QoS
Appreciating what QoS tools and methods do to packets in a network is fundamental to
appreciating the effect of individual QoS tools discussed in later chapters, and ultimately
understanding the network designs and methods that constitute the material in this book.
End-to-End QoS
Regardless of how it is measured, QoS is a vital element in any converged network. To
ensure the highest level of quality, however, QoS must be implemented across all areas of
the network.
QoS is described aptly by an age-old clich: Its only as strong as the weakest link.
Optimally, every device (host, server, switch, or router) that handles the packet along its
network path should employ QoS to ensure that the packet is not unduly delayed or lost
between the endpoints. It is a popular myth that QoS is applicable only to slow-speed widearea networks (WANs) or the Internet. Testing has shown that delay-sensitive applications,
such as voice, suffer quality loss when QoS is not enabled on campus devices, such as Cisco
IP phones and Cisco Catalyst switches.
Figure 1-3 shows the segments of the network where QoS must be deployed in the typical
IP telephony enterprise network and what QoS techniques are common in each segment.
Understanding QoS
Figure 1-3
11
Enabling QoS in an Enterprise Network

CallManager
cluster
PSTN
SRST
Router
Si
S
Si
Campus
Campus Access
Marking
Multiple queues
802.1p/Q
Fast link
convergence
Conditional trust
IP WAN
IP WAN
Bandwidth
Provisioning
Campus Distribution
Multiple queues
802.1p/Q
Classification
Reclassification
Branch Office
WAN Aggregation
Branch Router
Branch Switch
Multiple queues
802.1p/Q
Traffic Shaping
Link efficiency
(LFI, cRTP)
Classification
Reclassification
Multiple queues
802.1p/Q
Link efficiency
(LFI, cRTP)
Classification
Reclassification
Network-Base
Application
Recognition
Marketing
Multiple queues
802.1p/Q
All Packets Are (Not) Equal

Networks without QoS enabled are described as best-effort networks. Best-effort network
designs treat all packets as equally important. Such networks work well if there is enough
CPU, memory, and bandwidth to handle immediately all the packets traversing the network
as they arrive (in other words, at line rate). However, because this is rarely the case, the
following scenarios tend to emerge:
User-based contention occurs when packets from different users, user groups,
departments, or even enterprises contend for the same network resources.
Application-based contention occurs when different applications from the same user
or user group contend with each other for limited network resources. These
applications might have different service requirements from the network, as
illustrated in Figure 1-4.
12
Figure 1-4
Application-Based Contention
Data
Smooth/Bursty
Benign/Greedy
Largely Drop
Insensitive
Delay Insensitive
(Within Reason)
TCP Retransmits
Video
Voice
Smooth
Benign
Drop Sensitive
Delay Sensitive
UDP Best Priority
Bursty
Greedy
Drop Sensitive
Delay Sensitive
UDP Priority
To obtain the appropriate levels of service from often scarce network resources experiencing contention, the fundamental concept of QoS must be appliednamely, that all packets
are not equal. In other words, to resolve the contention, packets must be treated with managed unfairness (either preferentially or deferentially). In other words, administrative policies must be dened and deployed throughout the network infrastructure to ensure that each
node makes consistent decisions about the relative importance of an individual packet (or
ow) and adjusts its packet-handling behavior accordingly.
This concept of unfairness applies equally to trafc for which higher levels of priority must
be maintained (for example, to voice and video trafc) and trafc that is undesirable in the
network (denial-of-service [DoS] or worm-generated trafc, or even general web surng to
destinations unrelated to the goals of the enterprise). The latter category of trafc
introduces the concept of less than best-effort service, also referred to as Scavenger service
(based on an Internet 2 draft specication). Best-effort or better service is provided to all
desirable trafc on the network, but there is no sense in providing even best-effort service
to unwanted trafc. Such Scavenger trafc, if not dropped outright, is carried only when
spare capacity is available.
The Challenges of Converged Networks

The high-level goal of QoS technologies in a converged network is to abstract the fact that
there is only one network and to make voice, video, and data convergence appear transparent to the end users. To achieve this goal, QoS technologies allow different types of trafc
Understanding QoS
13
to contend inequitably for network resources. Real-time applications, such as voice or interactive video, can be given priority or preferential services over generic data applications
but not to the point that data applications are starving for bandwidth.
QoS is dened as the measure of a systems service availability and transmission quality.
Service availability is a crucial foundation element of QoS. Before any QoS can be implemented successfully, the network infrastructure should be designed to be highly available.
The target for high availability is 99.999 percent uptime, with only 5 minutes of downtime
permitted per year. The transmission quality of the network is determined by the following
factors:
Packet lossThis is a comparative measure of the number of packets faithfully

transmitted and received to the total number transmitted, expressed as a percentage.
Packet loss in the context of QoS does not relate to drops because of network outages
or link aps (because these are a function of a high-availability network design), but
instead relates to drops because of network congestion.
Delay (or latency)This is the nite amount of time that it takes a packet to reach
the receiving endpoint after being transmitted from the sending endpoint. In the case
of voice, this equates to the amount of time that it takes for speech to leave the
speakers mouth and be heard by the listeners ear. This time period is termed end-toend delay and is comprised of two components: xed network delay and variable
network delay.
In data networks carrying voice, xed network delay further is broken down
into the following:
Packetization delayThe time required to sample and encode voice or
video signals into packets.
Serialization delayThe time required to transmit the packet bits onto the
physical media, based on the clocking rate of the interface.
Propagation delayThe time required for the electrical/optical pulses to
traverse the media en route to their destination, based on the laws of physics.
Delay variation (or jitter)Also known as interpacket delay, this is the difference
in the end-to-end delay between sequential packets. For example, if one packet
requires 100 ms to traverse the network from the source endpoint to the destination
endpoint, and the following packet requires 125 ms to make the same trip, the delay
variation would be calculated as 25 ms, as illustrated in Figure 1-5.
14
Figure 1-5
Delay Variation
Sender
Receiver
Network
Sender Transmits
t
A
D1
D2 = D1
D3 = D 2
D4 = D 3
Receiver Receives
Each end station in a VoIP or video over IP conversation has a jitter buffer,
which is used to smooth out the variation in arrival times of data packets that
contain voice. Jitter buffers can be xed or adaptive.
Instantaneous changes in arrival times of packets that exceed the jitter
buffers capability to compensate result in jitter buffer overruns and
underruns.
A jitter buffer underrun occurs when the arrival times of packets increase to
the point that the jitter buffer has been exhausted and contains no packets
for the digital signal processors (DSP) to process when it is time to play the
next piece of voice or video, resulting in clips.
A jitter buffer overrun occurs when packets containing voice or video arrive
faster than the jitter buffer can accommodate. When this happens, packets
are dropped when it is time to play the voice or video samples, resulting in
degraded voice quality.
Variable network delay generally is caused by congestion within the
network.
QoS Models
As briey introduced earlier, there are two models for providing the differentiated levels of
network service required in converged networks: IntServ and DiffServ.
QoS Models
15
IntServ Overview
Think of best-effort IP service in terms of the regular mail (snail-mail) service. The mail is
delivered if and when it can be, but it also might be lost (arriving at some undetermined time
in the future or not at all). By comparison, IntServ is analogous to a custom mail service,
such as diplomatic mail or courier services, with certain parameters regarding loss and
delivery guaranteed.
More specically, IntServ is a specication of the following:
What the sender is sending (rate, maximum transmission unit [MTU], and so on), as
described by the Transmission Specication (TSpec)
What the receiver needs (bandwidth, MTU, and so on), as described by the Receiver
Specication (RSpec)
How the signaling is performed on the network by the sender and receiver (that is, the
use of RSVP)
The framework of IntServ preserves the end-to-end semantics of QoS for IP. Key endpoints
are the sender and receiver applications that request a desired service from the network for
a set of ows, as dened by the source address, destination address, transport protocol,
source port, and destination port.
IntServ describes three main classes of service that an application can request:
Guaranteed services (RFC 2212)Provides rm (mathematically provable) bounds

on end-to-end packet-queuing delays, making it possible to provide a service that
guarantees both delay and bandwidth
Controlled load (RFC 2211)Provides the application ow with a quality of service
closely approximating the QoS that the same ow would receive from an unloaded
network element, but uses capacity (admission) control to ensure that this service is
received even when the network element is overloaded
Best-effort serviceProvides no service guarantees of any type
The advantages of the IntServ model include the following:
Conceptual simplicity, facilitating the integration with network policy administration

Discrete per-ow QoS, making it architecturally suitable to voice calls
Call admission control (CAC) capabilities, which can indicate to endpoints whether
the desired bandwidth is available
The disadvantages of the IntServ model include these:
All network elements must maintain state and exchange signaling messages on a perow basis, which might require a signicant amount of bandwidth on large networks.
Periodic refresh messages are used, which might require protection from packet loss
to keep the session(s) intact.
All intermediate nodes must implement RSVP.
16
DiffServ Overview
Continuing the analogy of mail services, DiffServ can be compared to different tiers of mail
service, such as regular, registered, priority, and express mail. Each service offers particular
parameters of delivery, and the piece of mail is stamped at each intermediate handling point
to ensure that it gets the required service. However, there is no specic connection between
the sender of the piece of mail and the treatment that the mail receives.
The premise of DiffServ is very simple: It offers different network service levels to a
packet, thereby enable[ing] scalable service discrimination in the Internet without the
need for per-ow state and signaling at every hop (RFC 2474). Packets of a particular
service belong to a particular class, and the treatment of each class is described by
PHBs with which the network node must comply. Meaningful services can be constructed
by a combination of the following:
Setting a eld in the IP header upon network entry or at a network boundary (IPP
or DSCPs)
Using this eld to determine the nodes inside the network forward the packets
Conditioning the marked packets at network boundaries in accordance with the
requirements or rules of each class or service
The essence of DiffServ is the specication of PHBs that an application can receive from
the network:
Expedited forwarding (RFC 3246, previously RFC 2598)Provides a strictpriority service (compare this to express mail service)
Assured forwarding (RFC 2597)Provides a qualied delivery guarantee (compare

this to registered mail service) and makes the provision for oversubscription to this
service (specically, markdown and dropping schemes for excess trafc)
Class selectors (RFC 2474)Provides code points that can be used for backward
compatibility with IP Precedence models
Best-effort serviceProvides a delivery when possible service (compare this to

regular mail service)
DiffServ constructs services from the PHBs by classifying packets into classes at the edge,
or boundary, of the network and marking the packets accordingly. Optionally, the packets
can be metered with policing or shaping techniques. The core of the network implements
the PHBs and uses the packet markings to make queuing and dropping decisions.
The DiffServ model include these advantages:
ScalabilityNo state or ow information is required to be maintained.
17
PerformanceThe packet contents need be inspected only once for classication

purposes. At that time, the packet is marked and all subsequent QoS decisions are
made on the value of a xed eld in the packet header, reducing processing
requirements.
InteroperabilityAll vendors already are running IP.

FlexibilityThe DiffServ model does not prescribe any particular feature (such as a
queuing technique) to be implemented by a network node. The node can use whatever
features optimize its hardware and architecture, as long as it is consistent with the
behavior expectation dened in the PHBs.
The disadvantages of the DiffServ model include the following:
No end-to-end bandwidth reservations are present; therefore, guarantees of services

can be impaired by network nodes that do not implement the PHBs appropriately over
congested links or by nodes that are not engineered correctly for the expected trafc
volume of a specic class.
The lack of a per-ow/per-session CAC makes it possible for applications to congest

each other. (For example, if only enough bandwidth for 10 voice calls exists and an
11th call is permitted, all 11 calls suffer call-quality deterioration.)

A fair amount of literature exists on QoS concepts, technologies, and features. The purpose
of this book is not to reiterate in depth how these tools work, but rather to show how these
tools interact with each other and what combination of tools can be used to achieve the
overall design objectives of a network. However, to lay the context for the designs set forth
in this book, a brief overview of the toolset follows.
Generally, QoS tools fall into the following categories:
Classication and marking tools

Policing and shaping tools
Congestion-avoidance (selective dropping) tools
Congestion-management (queuing) tools
Link-specic tools
Figure 1-6 illustrates the relationship and overall cohesiveness of different QoS tools.
18
Figure 1-6
QoS Toolset
Ingress
Egress
LLQ
IP Traffic
VoIP HTTP FTP
Marker or
Policer
DSCP Written
VoIP HTTP FTP
Classifier
WRED
Classification and Marking
Queuing
and
Shaping
Congestion
Avoidance
Scheduling
Interface
Congestion
Management
Packets or frames entering a network device must be analyzed to determine the treatment
that they should be given. This analysis is referred to as classication. Classication is the
rst QoS function to occur for any given QoS policy, and it can occur repeatedly at various
stages of policy enforcement. To reduce the requirement for detailed recursive classication, it generally is recommended that trafc types be classied as close to their sources as
possible and that their packets be marked accordingly.
Marking establishes a distinct trust boundary that demarcates the point where the packet
markings are set properly and, therefore, detailed classication (Layer 4 through 7 analysis)
no longer is required.
When packets enter a network device, three generic marking possibilities exist:
Packets are unmarked.

Packets are marked, but the markings are not trusted.
Packets are marked and the markings are trusted.
In the rst two scenarios, it is recommended that the packets be marked or re-marked.
After marking, the next step is another classication process based on packet markings. At
this point, packets might be discarded by a policer (for example, if they exceed an SLA) or
a congestion-avoidance mechanism (for example, early dropping because buffers reached
Simplifying QoS
19
the maximum limit). Packets that are not discarded are subject to congestion management
(queuing) to prioritize and protect various trafc types when congestion happens to the
transmission link. Finally, these packets are scheduled for transmission on the egress link,
where shaping might occur to control bursts and ensure that outgoing trafc conforms to
any SLA that is in effect on the ingress of the next hop.
Link-specic tools usually are required only at WAN edges and include mechanisms for
compression or fragmentation to reduce delay and jitter.
Some QoS tools (such as marking and policing) can be applied in both the ingress and
egress directions of the trafc ow on an interface. Other tools (such as queuing) can be
applied only in the egress direction (with some platform-specic exceptions).
The QoS mechanisms (primarily DiffServ) described previously are applicable to packets
already admitted to the network. Such tools are very effective in protecting real-time (voice)
from non-real-time (data) trafc, but they are completely ineffective in protecting real-time
applications from other real-time applications (that is, protecting voice trafc from other
voice trafc). Such protection can be achieved only through call admission control (CAC)
mechanisms, which decide whether to allow or disallow new packet streams onto the
network. IntServ mechanisms, such as RSVP, can be used to implement discrete CAC.
Simplifying QoS
In the late 1990s, Cisco spent signicant development effort to implement an extensive QoS
tool and feature set. The portfolio of QoS mechanisms and options became very rich and,
simultaneously, very complicated to deploy. During the early 2000s, the predominant
customer feedback relating to QoS was the request to simplify QoS deployment. In
response to this feedback, a series of QoS simplication and automation projects was
initiated. The result included the following:
The creation of a Modular QoS Command-Line Interface (CLI)

The establishment of a QoS Baseline
Default behavior (conforming to the QoS Baseline)
Cross-platform feature consistency
Automatic QoS (AutoQoS)
Modular QoS Command-Line Interface

The rst step toward simplication was to modularize the conguration of QoS and make
it consistent across platforms. To this end, a new conguration syntax, termed Modular QoS
Command-Line Interface (MQC), was introduced in Cisco IOS Release 12.0(5)T.
20
Three main components make up the MQC:
class-mapA classication lter dened within the policy map to identify trafc for
preferential or deferential treatment. Trafc can be identied by IPP or DSCP, named
or numbered access control lists (ACLs), Network-Based Application Recognition
(NBAR), Layer 2 parameters (CoS, FR DE, ATM cell loss priority [CLP], MPLS
Experimental [EXP] value), or a combination of these.
policy-mapA statement that denes how each trafc type, as identied by the class
map(s), should be serviced. Options include marking/re-marking, policing, shaping,
low-latency or class-based weighted fair queuing, selective dropping, and header
compression.
service-policyA statement that binds the policy to an interface and species

direction.
Example 1-1 demonstrates a sample MQC policy. This is just an example of a possible
conguration; the signicance of the commands used in this example is explained in
subsequent chapters.
Example 1-1 MQC Policy
Router(config)# class-map match-any CRITICAL-DATA
Router(config-cmap)# match protocol http url "*customer*"
Router(config-cmap)# match protocol citrix
Router(config)# policy-map WAN-EDGE
Router(config-pmap)# class CRITICAL-DATA
Router(config-pmap-c)# bandwidth 1000
Router(config)# interface Serial 0/0
Router(config-if)# service-policy output WAN-EDGE
QoS Baseline
Although MQC was a major step in simplifying QoS features, it addressed only the user
interface element of QoS deployment. Another factor in the complexity of QoS deployment
is the inconsistency of QoS features across specic hardware platform releases or software
releases.
In 2002, a series of internal Cisco initiatives, dubbed Technology Baselines, was launched
to address the inconsistency of which features exist on which platforms. Specic to QoS, the
QoS Baseline is a strategic internal document written by Ciscos most qualied QoS experts,
each of whom had contributed to multiple QoS RFCs and so was best qualied to interpret
these standards in additional detail. The QoS Baseline was developed with two primary
goals:
To document which QoS features are required on platforms that are considered
QoS-enabled products
To provide a gap analysis of which features, deemed important by the baseline,

existed on what platforms
Simplifying QoS
21
In part, the Cisco QoS Baseline is aimed at producing higher-quality new products,
predictable QoS feature sets across products, and improved consistency of QoS features
across platforms. To that end, it provides engineers who are developing new products with
a reference list of QoS features that they must include and test. For existing platforms, it
provides a gap analysis that is used to steer product roadmaps to achieve QoS baseline
compliance.
Beyond its engineering inuence, the overall objective of the QoS Baseline is to unify QoS
within Cisco: from service provider to enterprise, from engineering to marketing. For
example, the concept of trafc classication inevitably begs the question of how many
classes of trafc there should be. MQC supports up to 256 different trafc classes within a
policy map. Although adept QoS administrators might see this as desirable, those who are
new to QoS and have no idea how many classes of trafc they need in their networks might
view it as daunting.
To address the needs of the both expert and casual QoS users, the QoS Baseline denes a
set of class template recommendations that can be either be implemented as is or customized to individual needs. It gives a starting point for network design and implementation,
and it provides a basis for comparable product testing and performance reportingboth of
which signicantly simplify the implementation of QoS in a network.
The QoS Baseline denes up to 11 classes of trafc that are used in all Cisco design guides,
conguration examples, and testing suites. The QoS Baseline does not dictate that every
enterprise deploy all these trafc classes immediately. The set of 11 classes is designed to
accommodate the QoS needs of an enterprise not only for today, but also for the foreseeable
future. Even if an enterprise needs only a handful of these 11 classes today, following the
QoS Baseline recommendations will enable it to smoothly implement additional trafc
classes in the future.
Table 1-1 gives a summary of the QoS Baseline recommendations.
Default Behavior
The next step in the trend toward simplication was for the network nodes to do the right
thing. Through a series of changes in Cisco IOS Software and selected nonCisco IOS
products (such as IP phones), voice packets now are marked by default to the recommended
value of DSCP EF. These adjustments in the default behavior eliminate the need for access
lists and explicit marking congurations that were required in earlier software releases.
These changes are discussed in more detail in Chapter 3, Classication and Marking
Tools.
QoS Baseline Recommendations
22
Table 1-1
DSCP
DSCP Binary Value
Reference
Intended Protocols
Configuration
EF
EF
101110
RFC 3246
Interactive voice
Admission control =
RSVP
Queuing = priority
AF1
AF2
AF3
AF4
AF11
AF12
AF13
001010
001100
001110
RFC 2597
AF21
AF22
AF23
010010
010100
010110
RFC 2597
AF31
AF32
AF33
011010
011100
011110
RFC 2597
AF41
AF42
AF43
100010
100100
100110
RFC 2597
Bulk transfers, web,

general data service
Queuing = rate based
Database access,
transaction services,
interactive trafc,
preferred data service
Locally dened;
mission-critical
applications
Interactive video and

associated voice
Admission control =
RSVP
Active queue
management =
DSCP-based WRED
Active queue
management =
DSCP-based WRED
Active queue
management =
DSCP-based WRED

Active queue
management =
DSCP-based WRED
IP routing
Class 6
110000
RFC 2474
BGP, OSPF, and so on
section 4.2.2

Small guaranteed
minimum rate
Active queue
management = WRED
PHB
Table 1-1
QoS Baseline Recommendations (Continued)

PHB
DSCP
DSCP Binary Value
Reference
Intended Protocols
Streaming video
Class
Selector 4
100000
RFC 2474
Often proprietary
section 4.2.2
Configuration
Admission control =
RSVP
Active queue
management = WRED
Telephony signaling Class

(voice and video)
Selector 3
011000
RFC 2474
SIP, H.323, and so on
section 4.2.2

Small guaranteed
minimum rate
Active queue
management = WRED
Network
management
Class
Selector 2
010000
RFC 2474
SNMP
section 4.2.2

Small guaranteed
minimum rate
Active queue
management = WRED
Scavenger
I2SS or
Class
Selector 1
001000
Internet 2
usage
User-selected service

No bandwidth
guarantee
Active queue
management = WRED
Other
000000
RFC 2474
section 4.1
Unspecied trafc

Minimal bandwidth
guarantee
Active queue
management or
Per-ow fair queuing
23
Active queue
management = WRED
Simplifying QoS
Default or
Class
Selector 0
24
Cross-Platform Feature Consistency

An additional objective in the simplication of QoS (beyond syntax and automation) is to
improve the operational and semantic feature consistency across platforms.
As part of this effort, Cisco is consolidating and improving the QoS code bases of the distributed router architectures (such as the Cisco 7500 series of routers) and the nondistributed router families (such as the Cisco 7200 to 1700 series routers). This initiative is called
Consistent QoS Behavior code and should remove most (if not all) of the traditional QoS
idiosyncrasies between the distributed and nondistributed platforms.
Such projects are ongoing, as the application drivers for QoS continue to demand unique
features for specic scenarios, but administrators generally prefer QoS consistency and
simplicity.
Automatic QoS
Automatic QoS (AutoQoS) is essentially an intelligent macro that enables an administrator
to enter one or two simple AutoQoS commands to enable all the appropriate features for the
recommended QoS settings for an application on a specic interface.
For example, in its rst release, AutoQoS VoIP would provide best-practice QoS congurations for VoIP on Cisco Catalyst switches and routers. By entering one global or one
interface command (depending on the platform), the AutoQoS VoIP macro then would
expand these commands into the recommended VoIP QoS congurations (complete with
all the calculated parameters and settings) for the platform and interface on which the AutoQoS is applied.
AutoQoS is available on both LAN and WAN Cisco Catalyst switches and Cisco IOS
routers. In its initial version, however, AutoQoS applies only to VoIP deployments.
For campus Catalyst switches, AutoQoS performs the following automatically:
Enforces a trust boundary at Cisco IP phones
Modies queue admission criteria (such as CoS-to-queue mapping)
Enforces a trust boundary on Catalyst switch access ports and uplinks/downlinks

Enables Catalyst strict priority queuing for voice and weighted round-robin queuing
for data trafc
Modies queue sizes and queue weights, where required
Modies CoS-to-DSCP and IP precedence-to-DSCP mappings
Simplifying QoS
25
For Cisco IOS routers, AutoQoS is supported on Frame Relay, ATM, HDLC, Point-to-Point
Protocol (PPP), and Frame Relay-to-ATM links. It performs the following automatically:
Classies and marks VoIP bearer trafc (to DSCP EF) and call-signaling trafc
(originally to DSCP AF31, but more recently to CS3)
Applies scheduling:
Low-latency queuing (LLQ) for voice
Class-based weighted fair queuing (CBWFQ) for call signaling
Fair queuing (FQ) for all other trafc
Enables Frame Relay trafc shaping with optimal parameters, if required
Enables IP RTP header compression (cRTP), if required
Enables Link Fragmentation and Interleaving (either MLPPP LFI or FRF.12) on slow
links (those less than or equal to 768 kbps), if required
Provides RMON alerts of VoIP packets that are dropped
The rst phase of AutoQoS on the Cisco IOS router platforms is available in Cisco IOS
Release 12.2(15)T.
In its second releasefor Cisco IOS routers onlyAutoQoS Enterprise detects and
provisions for up to 10 (of the 11 QoS Baseline) trafc classes. (The only class that is not
assigned automatically is Locally-Dened Mission-Critical because it is dened
subjectively by network administrators and varies from one enterprise to the next.)
The AutoQoS Enterprise feature simplies QoS implementation and speeds the
provisioning of QoS for voice, video, and data over a Cisco network. It reduces human error
and lowers training costs. AutoQoS Enterprise creates class maps and policy maps on the
basis of Cisco experience and best-practices methodology, such as that documented in this
book. Customers also can use existing Cisco IOS commands to modify the congurations
that automatically are generated by the AutoQoS Enterprise feature, to meet specic
requirements.
The AutoQoS Enterprise feature consists of two conguration phases, completed in the
following order:
1 Autodiscovery (data collection)The autodiscovery phase uses protocol discovery
based on Network-Based Application Recognition (NBAR) to detect the applications

on the network and perform statistical analysis on the network trafc.
2 AutoQoS template generation and installationThis phase generates templates
from the data collected during the autodiscovery phase and installs the templates on
the interface. These templates then are used as the basis for creating the class maps
and policy maps for the network. After the class maps and policy maps are created,
they are installed on the relevant interface(s).
26
Table 1-2 shows the classes of trafc automatically detected and provisioned for by
AutoQoS Enterprise, complete with their QoS Baseline recommended markings.
Table 1-2
AutoQoS Enterprise Class Denitions and Marking

AutoQoS Class
Name
Traffic Type
DSCP Value
IP Routing
Network-control trafc, such as routing protocols
CS6
Interactive Voice
Interactive voice-bearer trafc
EF
Interactive Video
Interactive video-data trafc
AF41
Streaming Video
Streaming media trafc
CS4
Telephony Signaling
Telephony signaling and control trafc
AF31 or CS3
Transactional/Interactive Database applications transactional in nature
AF21
Network Management
Network-management trafc
CS2
Bulk Data
Bulk data transfers, web trafc, general data

service
AF11
Scavenger
Casual entertainment, rogue trafc (trafc in this

category is given less-than best-effort treatment)
CS1
Best Effort
Default class, all noncritical trafc, HTTP, all

miscellaneous trafc
This second phase of AutoQoS became available in Cisco IOS Release 12.3(7)T.
If I Have AutoQoS, Why Should I Be Reading This

Book?
It is important to keep in mind not only the evolution of QoS, but also, more specically,
that of AutoQoS.
In the late 1990s, critical QoS features for enabling the convergence of voice and data
networks, such as low-latency queuing (LLQ), Link Fragmentation and Interleaving (LFI),
Catalyst queuing, and trust were developed and released within Cisco IOS and within
Catalyst hardware/software.
Despite the recognition that these features could enable network convergence, many
customers were slow to adopt and deploy these technologies. Mainly, this was because they
were at a loss for how best to deploy such QoS mechanisms to achieve the end-to-end
service levels that VoIP required. Many nerd-knobs within the QoS toolset could be
turned and tuned, and overall the tools were complex and disparate, and required an indepth understanding to be used effectively.
If I Have AutoQoS, Why Should I Be Reading This Book?
27
To address this barrier to adoption, a small group of dedicated technical marketing engineers within a Cisco solutions engineering team was tasked with guring out how best to
deploy QoS for converged voice and data networks and publishing a best-practices design
guide that customers could use as a reference. To accomplish this mandate, they built the
largest networking lab within Cisco to run extensive scale tests of QoS features on various
hardware and software platforms.
Unlike conventional regression testing of QoS features, this team used a different approach:
They tested QoS features not in isolation, but in conjunction with the many other features
that typically would be deployed within a large enterprise environment. For example, QoS
tools were enabled simultaneously with availability tools, multicast tools, and security and
encryption tools. The objective was to make the tests as representative of real-life enterprise
environments as possible.
The results from these extensive tests were summarized and published in late 1999 as the
QoS Design Guide. This document quickly became one of the most downloaded technical
documents ever published by Cisco. Customers nally had not only the tools to achieve
convergence of their voice and data networks, but also the veried design guidance to do
so with condence.
Nonetheless, at around 200 pages, the conguration-rich QoS Design Guide was a serious
read that took a long time for most network administrators to fully absorb. Furthermore, it
indirectly highlighted how difcult it was to enable end-to-end QoS across a single vendors
(Ciscos) products. There were far too many platform-specic idiosyncrasies to keep in
mind, as far as most network administrators were concerned.
Thus, to simplify the process, the solutions engineering team engaged with various
technology groups and business units within Cisco to drive the automation of these bestpractice recommendations. The response was very favorable, and the result was AutoQoS
VoIP for the campus and WAN. AutoQoS VoIP is a cross-platform feature that is targeted
to customers who want to deploy QoS for VoIP quickly and accurately, without the
requirement of extensive knowledge of QoS.
As with all technical papers, the QoS Design Guide needed to be updated and expanded
after a couple years. During this time, customers increasingly were struggling with how
best to deploy QoS for videoconferencing and different types of data applications. Therefore, the scope of the original design guide was expanded to include these applications, and
a second version of the document was released in August 2002. At the time, all design
guides originating from the solutions engineering team were rebadged as Solution Reference Network Designs (SRNDs), to set them apart from the many (often unveried and not
scale-tested) design guides available on Cisco.com. Thus, the second version of this document was titled the Enterprise QoS SRND.
Shortly thereafter, the QoS Baseline was completed and released internally. Because the
QoS Baseline conicted with some of the (DSCP marking) recommendations published in
the QoS SRND, it took precedence and required the Enterprise QoS SRND to be rewritten
28
to be compliant with the QoS Baseline. This caused changes in the document because the
marking recommendations put forward in the Enterprise QoS SRND reected the best
practices for an enterprise, but not necessarily for a service provider. As the lines between
enterprise and service provider are not only blurring but also requiring a level of cooperation previously unprecedented (as discussed in additional detail in Chapter 15, MPLS
VPN QoS Design), it is important for a single set of marking recommendations to be used
for both enterprise and service provider design guides.
Following this, Cisco IOS development combined the best practices put forward in the
QoS design guides with the classication and marking recommendations dened in the QoS
Baseline, and released AutoQoSEnterprise for the WAN. AutoQoS Enterprise, similar to
AutoQoS VoIP, is targeted at customers who want to enable QoS for voice, video, and
multiple classes of data quickly and accurately over their WANs, without requiring an
extensive knowledge of QoS conguration and operation.
So why is this relevant? This brief history was given to describe the cycles required in
AutoQoS evolution, illustrated in Figure 1-7.
Figure 1-7
AutoQoS Evolution
QoS SRND v3
(Voice, Video, Data +
DoS/Worm Mitigation)
Increased Sophistication
Advanced Data QoS Features

(Advanced Campus Policers)
QoS Baseline
Data QoS Features

(NBAR, DSCP-WRED)
VoIP QoS Features

(LLQ, LFI)
1998
QoS SRND v2
(Voice, Video, Data)
QoS SRND v1
(VoIP Only)
2000
AutoQoS VoIP
(Campus + WAN)
2002
Specically, AutoQoS evolution depends on two prerequisites:
QoS feature development

Veried network design guides
AutoQoS-Enterprise
(WAN Only)
2004
Summary
29
The Continuing Evolution of QoS

As a technology, QoS is still evolving, with new features continually being developed.
Likewise, the role of QoS is evolving.
Specically, following the development of advanced hardware policers within the Catalyst
switching platforms, such as per-port/per-VLAN policers, dual-rate policers, and userbased microow policers, new contexts for the use of QoS tools have become possible.
These include the use of QoS as a technology not only to enable convergence, but also to
increase securityespecially in relation to the exponentially increasing number of DoS and
worm attacks experienced by customers since 2000.
Therefore, this book, which represents the third generation of the Cisco QoS design guides,
details the role of QoS technologies in protecting voice, video, and critical data, and how
these tools can be used to mitigate DoS and worm attacks.
As QoS continues to evolve, it is the goal of the authors that as networks increasingly adopt
these recommended designs, they will be automated through a future release of AutoQoS.
Therefore, the value in reading this book is to learn the rationale behind the best practices
that went into AutoQoS VoIP and AutoQoS Enterprise development, and to go beyond these
contexts and see the next-generation strategic role of QoS technologies.
Summary
This chapter briey reviewed the fundamental concepts of QoS, such as IntServ and
DiffServ; the categories of QoS tools that exist, and why they are categorized in this way;
the Modular QoS CLI that pulls together the use and conguration of various QoS tools,
making it clearer how and where they interact; the QoS Baseline; and Automatic QoS Cisco
initiatives. This chapter also introduced the QoS Baseline 11 classes of trafc, including the
less-than best-effort Scavenger class.
You are now ready to get into the principles of QoS design and then review salient details
of the QoS toolset. Chapter 2, QoS Design Overview, introduces QoS design principles
and objectives. Chapters 3 through 9 provide a more in-depth discussion of the various QoS
tool categories, the specic tools in each, and how and where these should be applied.
Chapters 10 through 16 provide design examples of how these tools are deployed in specic
types of networks. These network types are included in these discussions:
LAN QoS design

WAN QoS design, covering both remote ofce and WAN aggregator designs
MPLS and IPSec VPN QoS design
30
The objective of this book is not to discuss every QoS feature in isolationplenty of
material available in the industry accomplishes this. The goal instead is to show how these
techniques are used in a holistic network design to achieve end-to-end QoS for the entire
network. Another objective of this book is to show how to deploy QoS not only for purposes
such as maintaining voice quality for VoIP calls, but also to show how QoS can be used to
achieve such goals as protecting the network against denial-of-service (DoS) attacks and
managing trafc ows to keep unwanted trafc (such as copyright-infringing music lesharing applications) off the network.
Further Reading
General
IP Quality of Service. Srinivas Vegesna. Indianapolis: Cisco Press, 2001.
Cisco DQOS Exam Certication Guide. Wendell Odom, Michael Cavanaugh.

Indianapolis: Cisco Press, 2003.
Communications Convergence.com: http://www.cconvergence.com.
Cisco.com QoS page: http://www.cisco.com/go/qos.
IntServ is described by a series of RFCs from the IntServ IETF working group.
RFC 2205, Resource Reservation Protocol (RSVP) Version 1 Functional

Specication: http://www.ietf.org/rfc/rfc2205.txt.
RFC 2210, The Use of RSVP with IETF Integrated Services: http://www.ietf.org/
rfc/rfc2210.txt.
RFC 2211, Specication of the Controlled-Load Network Element Service: http://

www.ietf.org/rfc/rfc2211.txt.
RFC 2212, Specication of Guaranteed Quality of Service: http://www.ietf.org/rfc/

rfc2212.txt.
RFC 2215, General Characterization Parameters for Integrated Service Network

Elements: http://www.ietf.org/rfc/rfc2215.txt.
Cisco Catalyst QoS: Quality of Service in Campus Networks. Michael Flannagan and
Kevin Turek. Indianapolis: Cisco Press, 2003.
CommWeb Tech Library: http://techlibrary.commweb.com/data/web/cweb/

cweb_index.jsp.
IntServ
RFC 1633, Integrated Services in the Internet Architecture: An Overview: http://
Further Reading
31
DiffServ
DiffServ is described by a series of RFCs from the DiffServ IETF working group.
RFC 2474, Denition of the Differentiated Services Field (DS Field) in the IPv4 and
IPv6 Headers: http://www.ietf.org/rfc/rfc2474.txt.
RFC 2475, An Architecture for Differentiated Services: http://www.ietf.org/rfc/
rfc2475.txt.
RFC 2597, Assured Forwarding PHB Group: http://www.ietf.org/rfc/rfc2597.txt.
RFC 2598, An Expedited Forwarding PHB: http://www.ietf.org/rfc/rfc2598.txt.
RFC 2697, A Single Rate Three Color Marker: http://www.ietf.org/rfc/rfc2697.txt.
RFC 2698, A Two Rate Three Color Marker: http://www.ietf.org/rfc/rfc2698.txt.
RFC 3168, Explicit Congestion Notication (ECN) to IP: http://www.ietf.org/rfc/
rfc3168.txt.
RFC 3246, An Expedited Forwarding PHB (replacing RFC 2598): http://
AutoQoS
AutoQoS VoIP for Cisco IOS router platforms, Cisco IOS Release 12.2(15)T
documentation: http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/
122newft/122t/122t15/ftautoq1.htm.
AutoQoS Enterprise for Cisco IOS router platforms, Cisco IOS Release 12.3(7)T
documentation: http://www.cisco.com/univercd/cc/td/doc/product/software/ios123/
123newft/123t/123t_7/ftautoq2.htm.
AutoQoS VoIP, Catalyst 2950, Catalyst IOS version 12.1(19)EA1: http://
www.cisco.com/univercd/cc/td/doc/product/lan/cat2950/12119ea1/2950scg/
swqos.htm#1125412.
swqos.htm#1231112.
www.cisco.com/univercd/cc/td/doc/product/lan/c3550/12119ea1/3550scg/
swqos.htm#1185065.
swqos.htm#1231112.
AutoQoS VoIP, Catalyst 4500, Catalyst IOS version 12.2(18): http://www.cisco.com/
univercd/cc/td/doc/product/lan/cat4000/12_2_18/cong/qos.htm#1281380.
AutoQoS VoIP, Catalyst 6500, Catalyst CatOS version 8.2: http://www.cisco.com/
univercd/cc/td/doc/product/lan/cat6000/sw_8_2/confg_gd/autoqos.htm.
This chapter provides an overview of the QoS design and deployment process. This process
requires business-level objectives of the QoS implementation to be dened clearly and for
the service-level requirements of applications to be assigned preferential or deferential
treatment so that they can be analyzed.
These enterprise applications with unique QoS requirements are discussed in this chapter:
Voice
Call-Signaling
Interactive-Video
Streaming-Video
Best-Effort Data
Bulk Data
Transactional Data
Mission-Critical Data
IP Routing trafc
Network-Management trafc
Scavenger trafc
Additionally, key QoS design and deployment best practices that can simplify and expedite
QoS implementations are presented, including these:
Classication and marking principles

Policing and markdown principles
Queuing and dropping principles
DoS and worm mitigation principles
Deployment principles
CHAPTER
QoS Design Overview

More than just a working knowledge of QoS tools and syntax is needed to deploy end-toend QoS in a holistic manner. First, it is vital to understand the service-level requirements
of the various applications that require preferential (or deferential) treatment within the
network. Additionally, a number of QoS design principles that extensive lab testing and
customer deployments have helped shape can streamline a QoS deployment and increase
the overall cohesiveness of service levels across multiple platforms.
This chapter overviews the QoS requirements of VoIP, Video (both Interactive-Video
and Streaming-Video), and multiple classes of data. Within this discussion, the QoS
requirements of the control plane (routing and management trafc) are considered. The
Scavenger class is examined in more detail, and a strategy for mitigating DoS and worm
attacks is presented.
Next, QoS design principles relating to classication, marking, policing, queuing, and
deployment are discussed. These serve as guiding best practices in the design chapters to
follow.
QoS Requirements of VoIP

VoIP deployments require the provisioning of explicit priority servicing for VoIP (bearer
stream) trafc and a guaranteed bandwidth service for Call-Signaling trafc. These related
classes are examined separately.
Voice (Bearer Trafc)

The following list summarizes the key QoS requirements and recommendations for voice
(bearer trafc):
Voice trafc should be marked to DSCP EF per the QoS Baseline and RFC 3246.
Loss should be no more than 1 percent.
One-way latency (mouth to ear) should be no more than 150 ms.
Average one-way jitter should be targeted at less than 30 ms.
A range of 21 to 320 kbps of guaranteed priority bandwidth is required per call
(depending on the sampling rate, the VoIP codec, and Layer 2 media overhead).
34
Chapter 2: QoS Design Overview
Voice quality directly is affected by all three QoS quality factors: loss, latency, and jitter.
Loss
Loss causes voice clipping and skips. Packet loss concealment (PLC) is a technique used
to mask the effects of lost or discarded VoIP packets. The method of PLC used depends
upon the type of codec. A simple method used by waveform codecs such as G.711 (PLC
for G.711 is dened in G.711 Appendix I) is to replay the last received sample with
increasing attenuation at each repeat; the waveform does not change much from one sample
to the next. This technique can be effective at concealing the loss of up to 20 ms of samples.
The packetization interval determines the size of samples contained within a single packet.
Assuming a 20-ms (default) packetization interval, the loss of two or more consecutive
packets results in a noticeable degradation of voice quality. Therefore, assuming a random
distribution of drops within a single voice ow, a drop rate of 1 percent in a voice stream
would result in a loss that could not be concealed every 3 minutes, on average. A 0.25
percent drop rate would result in a loss that could not be concealed once every 53 minutes,
on average.
NOTE
A decision to use a 30-ms packetization interval, for a given probability of packet loss,
could result in worse perceived call quality than for 20 ms because PLC could not
effectively conceal the loss of a single packet.
Low-bit-rate, frame-based codecs, such as G.729 and G.723, use more sophisticated PLC
techniques that can conceal up to 30 to 40 ms of loss with tolerable quality when the
available history used for the interpolation is still relevant.
With frame-based codecs, the packetization interval determines the number of frames
carried in a single packet. As with waveform-based codecs, if the packetization interval is
greater than the loss that the PLC algorithm can interpolate for, PLC cannot effectively
conceal the loss of a single packet.
VoIP networks typically are designed for very close to 0 percent VoIP packet loss, with the
only actual packet loss being due to L2 bit errors or network failures.
Latency
Latency can cause voice quality degradation if it is excessive. The goal commonly used in
designing networks to support VoIP is the target specied by ITU standard G.114 (which,
incidentally, is currently under revision): This states that 150 ms of one-way, end-to-end
(from mouth to ear) delay ensures user satisfaction for telephony applications. A design
35
should apportion this budget to the various components of network delay (propagation
delay through the backbone, scheduling delay because of congestion, and access link
serialization delay) and service delay (because of VoIP gateway codec and dejitter buffer).
Figure 2-1 illustrates these various elements of VoIP latency (and jitter because some delay
elements are variable).
Figure 2-1
Elements Affecting VoIP Latency and Jitter
Si
Si
WAN/VPN
Campus
CODEC
G.729A: 25 ms
Branch
O
Offic
ffic
ffice
Office
Queuing
Serialization
Propagation and
Network
Jitter Buffer
Variable
Variable
Fixed
(6.3 s / Km) +
Network Delay
(Variable)
20-50 ms
End-to-End Delay Must Be
150ms
If the end-to-end voice delay becomes too long, the conversation begins to sound like two
parties talking over a satellite link or even a CB radio. The ITU G.114 states that a 150-ms
one-way (mouth-to-ear) delay budget is acceptable for high voice quality, but lab testing
has shown that there is a negligible difference in voice quality mean opinion scores (MOS)
using networks built with 200-ms delay budgets. Thus, Cisco recommends designing to the
ITU standard of 150 ms. If constraints exist and this delay target cannot be met, the delay
boundary can be extended to 200 ms without signicant impact on voice quality.
NOTE
Certain organizations might view higher delays as acceptable, but the corresponding
reduction in VoIP quality must be taken into account when making such design decisions.
36
Jitter
Jitter buffers (also known as playout buffers) are used to change asynchronous packet
arrivals into a synchronous stream by turning variable network delays into constant delays
at the destination end systems. The role of the jitter buffer is to trade off between delay and
the probability of interrupted playout because of late packets. Late or out-of-order packets
are discarded.
If the jitter buffer is set either arbitrarily large or arbitrarily small, it imposes unnecessary
constraints on the characteristics of the network. A jitter buffer set too large adds to the endto-end delay, meaning that less delay budget is available for the network; hence, the network needs to support a tighter delay target than practically necessary. If a jitter buffer is
too small to accommodate the network jitter, buffer underows or overows can occur.
In an underow, the buffer is empty when the codec needs to play out a sample. In an overow, the jitter buffer is already full and another packet arrives; that next packet cannot be
enqueued in the jitter buffer. Both jitter buffer underows and overows cause voice quality
degradation.
Adaptive jitter buffers aim to overcome these issues by dynamically tuning the jitter buffer
size to the lowest acceptable value. Well-designed adaptive jitter buffer algorithms should
not impose any unnecessary constraints on the network design by doing the following:
Instantly increasing the jitter buffer size to the current measured jitter value following
a jitter buffer overow
Slowly decreasing the jitter buffer size when the measured jitter is less than the current
jitter buffer size
Using PLC to interpolate for the loss of a packet on a jitter buffer underow
When such adaptive jitter buffers are usedin theoryyou can engineer out explicit
considerations of jitter by accounting for worst-case per-hop delays. Advanced formulas
can be used to arrive at network-specic design recommendations for jitter (based on
maximum and minimum per-hop delays). Alternatively, because extensive lab testing has
shown that voice quality degrades signicantly when jitter consistently exceeds 30 ms, this
30 ms value can be used as a jitter target.
Because of its strict service-level requirements, VoIP is well suited to the expedited
forwarding per-hop behavior, dened in RFC 3246 (formerly RFC 2598). Therefore, it
should be marked to DSCP EF (46) and assigned strict-priority servicing at each node,
regardless of whether such servicing is done in hardware (as in Catalyst switches through
1PxQyT queuing, discussed in more detail in Chapter 10, Catalyst QoS Tools) or in
software (as in Cisco IOS routers through LLQ, discussed in more detail in Chapter 5,
Congestion-Management Tools).
The bandwidth that VoIP streams consume (in bits per second) is calculated by adding the
VoIP sample payload (in bytes) to the 40-byte IP, UDP, and RTP headers (assuming that
cRTP is not in use), multiplying this value by 8 (to convert it to bits), and then multiplying
again by the packetization rate (default of 50 packets per second).
37
Table 2-1 details the bandwidth per VoIP ow (both G.711 and G.729) at a default packetization rate of 50 packets per second (pps) and at a custom packetization rate of 33 pps. This
does not include Layer 2 overhead and does not take into account any possible compression
schemes, such as Compressed Real-Time Transport Protocol (cRTP, discussed in detail in
Chapter 7, Link-Specic Tools).
For example, assume a G.711 VoIP codec at the default packetization rate (50 pps). A new
VoIP packet is generated every 20 ms (1 second / 50 pps). The payload of each VoIP packet
is 160 bytes; with the IP, UDP, and RTP headers (20 + 8 + 12 bytes, respectively) included,
this packet become 200 bytes in length. Converting bits to bytes requires multiplying by 8
and yields 1600 bps per packet. When multiplied by the total number of packets per second
(50 pps), this arrives at the Layer 3 bandwidth requirement for uncompressed G.711 VoIP:
80 kbps. This example calculation corresponds to the rst row of Table 2-1.
Table 2-1
NOTE
Voice Bandwidth (Without Layer 2 Overhead)

Bandwidth
Consumption
Packetization
Interval
Voice Payload
in Bytes
Packets Per
Second
Bandwidth Per
Conversation
G.711
20 ms
160
50
80 kbps
G.711
30 ms
240
33
74 kbps
G.729A
20 ms
20
50
24 kbps
G.729A
30 ms
30
33
19 kbps
The Service Parameters menu in Cisco CallManager Administration can be used to adjust
the packet rate. It is possible to congure the sampling rate above 30 ms, but this usually
results in poor voice quality.
A more accurate method for provisioning VoIP is to include the Layer 2 overhead, which
includes preambles, headers, ags, CRCs, and ATM cell padding. The amount of overhead
per VoIP call depends on the Layer 2 media used:
802.1Q Ethernet adds (up to) 32 bytes of Layer 2 overhead (when preambles are
included).
Point-to-Point Protocol (PPP) adds 12 bytes of Layer 2 overhead.
ATM adds varying amounts of overhead, depending on the cell padding requirements.
Multilink PPP (MLP) adds 13 bytes of Layer 2 overhead.

Frame Relay adds 4 bytes of Layer 2 overhead; Frame Relay with FRF.12 adds
8 bytes.
38
Table 2-2 shows more accurate bandwidth-provisioning guidelines for voice because it
includes Layer 2 overhead.
Table 2-2
Voice Bandwidth (Including Layer 2 Overhead)
Bandwidth
Consumption
802.1Q
Ethernet
PPP
MLP
Frame
Relay with
FRF.12
G.711 at 50 pps
93 kbps
84 kbps
86 kbps
84 kbps
106 kbps
G.711 at 33 pps
83 kbps
77 kbps
78 kbps
77 kbps
84 kbps
G.729A at 50 pps 37 kbps
28 kbps
30 kbps
28 kbps
43 kbps
G.729A at 33 pps 27 kbps
21 kbps
22 kbps
21 kbps
28 kbps
ATM
Call-Signaling Trafc
The following list summarizes the key QoS requirements and recommendations for CallSignaling trafc:
Call-Signaling trafc should be marked as DSCP CS3 per the QoS Baseline (during
migration, it also can be marked the legacy value of DSCP AF31).
150 bps (plus Layer 2 overhead) per phone of guaranteed bandwidth is required for
voice control trafc; more may be required, depending on the Call-Signaling
protocol(s) in use.
Originally, Cisco IP Telephony equipment marked Call-Signaling trafc to DSCP AF31.

However, the assured forwarding classes, as dened in RFC 2597, were intended for ows
that could be subject to markdown and aggressive dropping of marked-down values.
Marking down and aggressively dropping Call-Signaling could result in noticeable delay
to dial tone (DDT) and lengthy call-setup times, both of which generally translate into poor
user experiences.
Therefore, the QoS Baseline changed the marking recommendation for Call-Signaling
trafc to DSCP CS3 because Class-Selector code points, dened in RFC 2474, are not
subject to such markdown and aggressive dropping as Assured Forwarding Per-Hop
Behaviors are.
Some Cisco IP Telephony products already have begun transitioning to DSCP CS3 for CallSignaling marking. In this interim period, both code points (CS3 and AF31) should be
reserved for Call-Signaling marking until the transition is complete.
Most Cisco IP Telephony products use the Skinny Call-Control Protocol (SCCP) for CallSignaling. Skinny is a relatively lightweight protocol and, as such, requires only a minimal
amount of bandwidth protection (most of the Cisco large-scale lab testing was done by
39
provisioning only 2 percent for Call-Signaling trafc over WAN and VPN links). However,
newer versions of CallManager and SCCP have shown some bloating in this signaling
protocol, so design recommendations have been adjusted to match (most examples in the
design chapters that follow have been adjusted to allocate 5 percent for Call-Signaling
trafc). This is a normal part of QoS evolution: As applications and protocols continue to
evolve, so do the QoS designs required to accommodate them.
Other Call-Signaling protocols include (but are not limited to) H.225 and H.245, the
Session Initiated Protocol (SIP), and the Media Gateway Control Protocol (MGCP). Each
Call-Signaling protocol has unique TCP and UDP ports and trafc patterns that should be
taken into account when provisioning QoS policies for them.

Two main types of video trafc exist: Interactive-Video (videoconferencing) and
Streaming-Video (both unicast and multicast). Each type of video is examined separately.
Interactive-Video
When provisioning for Interactive-Video (video conferencing) trafc, the following
guidelines are recommended:
Interactive-Video trafc should be marked to DSCP AF41; excess videoconferencing

trafc can be marked down by a policer to AF42 or AF43.

One-way latency should be no more than 150 ms.
Jitter should be no more than 30 ms.
Assign Interactive-Video to either a preferential queue or a second priority queue
(when supported); when using Cisco IOS LLQ, overprovision the minimum-priority
bandwidth guarantee to the size of the videoconferencing session plus 20 percent.
(For example, a 384-kbps videoconferencing session requires 460 kbps of guaranteed
priority bandwidth.)
Because IP videoconferencing (IP/VC) includes a G.711 audio codec for voice, it has the
same loss, delay, and delay-variation requirements as voicebut the trafc patterns of
videoconferencing are radically different from those of voice.
For example, videoconferencing trafc has varying packet sizes and extremely variable
packet rates. These are illustrated in Figures 2-2 and 2-3.
40
Figure 2-2
Videoconferencing Trafc Packet-Size Breakdown

10251500 Bytes
37%
65128 Bytes
1%
129256 Bytes
34%
5131024 Bytes
20%
257512 Bytes
8%
Figure 2-3
Videoconferencing Trafc Rates (384-kbps Session Example)

I Frame
10241518
Bytes
I Frame
10241518
Bytes
450 kbps
30pps
P and B Frames
128256 Bytes
15pps
32 kbps
I frame is a full sample of the video.

P and B frames use quantization via
motion vectors and prediction algorithms.
The videoconferencing rate is the sampling rate of the video stream, not the actual bandwidth that the video call requires. In other words, the data payload of videoconferencing
packets is lled with 384 kbps of voice plus video samples. IP, UDP, and RTP headers
(40 bytes per packet, uncompressed) need to be included in IP/VC bandwidth provisioning,
as does the Layer 2 overhead of the media in use. Because (unlike VoIP) IP/VC packet sizes
and rates vary, the header overhead percentage also varies, so an absolute value of overhead
41
cannot be calculated accurately for all streams. However, testing has shown that a conservative rule of thumb for IP/VC bandwidth provisioning is to assign an LLQ bandwidth
equivalent to the IP/VC rate plus 20 percent. For example, a 384-kbps IP/VC stream
adequately is provisioned with an LLQ of 460 kbps.
NOTE
The Cisco LLQ algorithm has been implemented to include a default burst parameter
equivalent to 200 ms of trafc. Testing has shown that this burst parameter does not require
additional tuning for a single IP videoconferencing (IP/VC) stream. For multiple streams,
this burst parameter can be increased as required.
Streaming-Video
When addressing the QoS needs of Streaming-Video trafc, the following guidelines are
recommended:
Streaming-Video (whether unicast or multicast) should be marked to DSCP CS4, as

designated by the QoS Baseline.
There are no signicant jitter requirements.
Streaming-Video is typically unidirectional; therefore, remote branch routers might

not require provisioning for Streaming-Video trafc on their WAN or VPN edges (in
the direction of branch to campus).
Nonorganizational Streaming-Video applications (either unicast or multicast), such as

entertainment video content, may be marked as ScavengerDSCP CS1, provisioned
in the Scavenger trafc class and assigned a minimal bandwidth (CBWFQ)
percentage. For more information, see the Scavenger Class section, later in this
chapter.
Latency should be no more than 4 to 5 seconds (depending on the video applications

buffering capabilities).
Guaranteed bandwidth (CBWFQ) requirements depend on the encoding format and
rate of the video stream.
Streaming-Video applications have more lenient QoS requirements because they are not
delay sensitive (the video can take several seconds to cue up) and are largely not jitter
sensitive (because of application buffering). However, Streaming-Video might contain
valuable content, such as e-learning applications or multicast company meetings, in which
case it requires service guarantees.
The QoS Baseline recommendation for Streaming-Video marking is DSCP CS4.
42
An interesting consideration with respect to Streaming-Video comes into play when

designing WAN and VPN edge policies on branch routers: Because Streaming-Video is
generally unidirectional, a separate class likely is not needed for this trafc class in the
branch-to-campus direction of trafc ow.
Nonorganizational video content (or video thats strictly entertainment oriented in nature,
such as movies, music videos, humorous commercials, and so on) might be considered for
Scavenger service, meaning that these streams will play if bandwidth exists, but they will
be the rst to go during periods of congestion.
QoS Requirements of Data

Hundreds of thousands of data applications exist on the Internet, in all shapes and sizes.
Some are TCP, others are UDP; some are delay sensitive, others are not; some are bursty
in nature, others are steady; some are lightweight, others are bandwidth hogsthe list
goes on.
Data trafc characteristics vary from one application to another, as illustrated in Figure 2-4,
which compares an enterprise resource planning (ERP) application (Oracle) with another
(SAP).
Figure 2-4
Data Application Differences
Oracle
SAP R/3
064 Bytes
65127 Bytes
128252 Bytes
253511
Bytes
5121023
Bytes
10241518
Bytes
5121023
Bytes
064
Bytes
253511
Bytes
10241518
Bytes
128252
Bytes
65127
Bytes
43
To make the matter even more complicated, it is crucial to recognize that, just as applications vary one from another, even the same application can vary signicantly from one
version to another.
A brief anecdote speaks to this point: After a southern California semiconductor company
provisioned QoS for Voice and Mission-Critical Data (SAP R/3), everything went well for
about six months. At that point, users began complaining of excessive delays in completing
basic transactions. Operations that previously required a second or less to complete were
taking signicantly longer. The application teams blamed the networking teams, claiming
that QoS was broken. Further investigation produced the information in the following
graph, shown in Figure 2-5.
Figure 2-5
Data Application Version Differences
SAP Sales Order

Entry Transaction
VA01
# of
Bytes
SAP GUI Release 3.0 F
14,000
SAP GUI Release 4.6C, No Cache
57,000
SAP GUI Release 4.6C, with Cache
33,000
SAP GUI for HTML, Release 4.6C
490,000
Bytes per Transaction
Client Version
500,000
400,000
300,000
200,000
100,000
0
SAP GUI, SAP GUI,

Release
Release
3.0F
4.6C, with
Cache
SAP GUI, SAP GUI

Release (HTML),
4.6C, No Release
Cache
4.6C
The Mission-Critical Data applicationin this instance, SAPhad been upgraded from
version 3.0F to 4.6C. As a result, a basic order-entry transaction required 35 times more
trafc than the original version. Additional provisioning and policy tuning was required to
accommodate the new version of the same application.
Given this reality, the question on how best to provision QoS for data is a daunting one.
After wrestling with this question for several years, the authors of the QoS Baseline came
up with four main classes of data trafc, according to their general networking characteristics and requirements. These classes are Best-Effort, Bulk Data, Transactional Data/Interactive Data and (Locally-Dened) Mission-Critical Data. Each of these classes is examined
in more detail in the following sections.
44
Best-Effort Data
When addressing the QoS needs of Best-Effort trafc, the following guidelines are
recommended:
Best-Effort trafc should be marked to DSCP 0.

Adequate bandwidth should be assigned to the Best-Effort class as a whole because
the majority of applications default to this class. It is recommended to reserve at least
25 percent for Best-Effort trafc.
The Best-Effort class is the default class for all data trafc. Only if an application has been
selected for preferential or deferential treatment is it removed from the default class.
In 2003, one Wall Street nancial company did an extensive study to identify and categorize
the number of different applications on its networks. It found more than 3000 discrete applications traversing its infrastructure. Further research has shown that this is not uncommon for
larger enterprises. Therefore, because enterprises have several hundredif not thousands
ofdata applications running over their networks (of which the majority default to the
Best-Effort class), adequate bandwidth needs to be provisioned for this default class to handle the sheer volume of applications that are included in it. Otherwise, applications that
default to this class easily are drowned out, typically resulting in an increased number of
calls to the networking help desk from frustrated users. It is therefore recommended that at
least 25 percent of a links bandwidth be reserved for the default Best-Effort class.
Bulk Data
When addressing the QoS needs of Bulk Data trafc, the following guidelines are
recommended:
Bulk Data trafc should be marked to DSCP AF11; excess Bulk Data trafc can be
marked down by a policer to AF12 or AF13.
Bulk Data trafc should have a moderate bandwidth guarantee but should be
constrained from dominating a link.
The Bulk Data class is intended for applications that are relatively noninteractive and
not drop sensitive, and that typically span their operations over a long period of time
as background occurrences. Such applications include FTP, e-mail, backup operations,
database synchronizing or replicating operations, video content distribution, and any other
type of application in which users typically cannot proceed because they are waiting for the
completion of the operation (in other words, a background operation).
The advantage of provisioning moderate bandwidth guarantees to Bulk Data applications
(instead of applying policers to them) is that Bulk Data applications dynamically can take
advantage of unused bandwidth and thus can speed up their operations during nonpeak
periods. This, in turn, reduces the likelihood that they will bleed into busy periods and
absorb inordinate amounts of bandwidth for their non-time-sensitive operations.
45
Transactional Data/Interactive Data

When addressing the QoS needs of Transactional Data and Interactive Data trafc, the
following guidelines are recommended:
Transactional Data trafc should be marked to DSCP AF21; excess Transactional

Data trafc can be marked down by a policer to AF22 or AF23.
Transactional Data trafc should have an adequate bandwidth guarantee for the
interactive, foreground operations that it supports.
The Transactional Data/Interactive Data class is a combination of two similar types

of applications: Transactional Data client/server applications and interactive messaging
applications. For the sake of simplicity, this class is referred to as Transactional Data only.
The response-time requirement separates Transactional Data client/server applications
from generic client/server applications. For example, with Transactional Data client/server
applications (such as SAP, PeopleSoft, and Oracle), the user waits for the operation to
complete before proceeding (in other words, the transaction is a foreground operation).
E-mail is not considered a Transactional Data client/server application because most e-mail
operations happen in the background, and users usually do not notice even delays of several
hundred milliseconds in mailspool operations.
Locally Dened Mission-Critical Data

When addressing the QoS needs of Locally-Dened Mission-Critical Data trafc, the
following guidelines are recommended:
Locally-Dened Mission-Critical Data trafc should be marked to DSCP AF31;

excess Mission-Critical Data trafc can be marked down by a policer to AF32 or
AF33. However, Cisco IP Telephony equipment currently is using DSCP AF31 to
mark Call-Signaling trafc; until all Cisco IPT products mark Call-Signaling to DSCP
CS3, a temporary placeholder code point, DSCP 25, can be used to identify LocallyDened Mission-Critical Data trafc.
Locally-Dened Mission-Critical Data trafc should have an adequate bandwidth

guarantee for the interactive, foreground operations that it supports.
The Locally-Dened Mission-Critical class is probably the most misunderstood class

specied in the QoS Baseline. Under the QoS Baseline model, all trafc classes (with the
exclusion of Scavenger and Best-Effort) are considered critical to the enterprise. The
term locally dened is used to underscore the purpose of this class: for each enterprise to
have a premium class of service for a select subset of its Transactional Data applications
that have the highest business priority for it.
For example, an enterprise might have properly provisioned Oracle, SAP, BEA, and Siebel
within its Transactional Data class. However, the majority of its revenue might come from
SAP, so it might want to give this Transactional Data application an even higher level of
46
preference by assigning it to a dedicated class (such as the Locally-Dened MissionCritical class).

Because the admission criteria for this class is nontechnical (being determined by business
relevance and organizational objectives), the decision about which application(s) should be
assigned to this special class easily can become an organizationally and politically charged
debate. It is recommended to assign as few applications to this class (from the Transactional
Data class) as possible. In addition, it is recommended that executive endorsement for
application assignments to the Locally-Dened Mission-Critical class be obtained: The
potential for QoS deployment derailment exists without such an endorsement.
For the sake of simplicity, this class is referred to simply as Mission-Critical Data.
Based on these denitions, Table 2-3 shows some applications and their generic networking
characteristics, which determine what data application class they are best suited to.
Table 2-3
Data Applications by Class

Application
Class
Interactive
PHB: AF2
Transactional
PHB: AF2
Application/Traffic
Properties
Packet/Message
Sizes
Telnet, Citrix, Oracle

Thin-Clients, AOL Instant
Messenger, Yahoo! Instant
Messenger, PlaceWare
(Conference), Netmeeting
Whiteboard.
Highly interactive
applications with tight userfeedback requirements.
Average message
size < 100 bytes.
SAP, PeopleSoftVantive,
OracleFinancials,
Internet Procurement, B2B,
Supply Chain Management,
Application Server, Oracle
8i Database, Ariba Buyer,
I2, Siebel, E.piphany,
Broadvision, IBM Bus 2
Bus, Microsoft SQL, BEA
Systems, DLSw+.
Transactional applications
typically use a client/server
protocol model.
Depends on
application; could
be anywhere from
1 KB to 50 MB.
Example Applications
User-initiated, client-based
queries are followed by
server response. The query
response can consist of
many messages between
client and server.
The query response can
consist of many TCP and
FTP sessions running
simultaneously (for
example, HTTP-based
applications).
Max message size

< 1 KB.
Table 2-3
47
Data Applications by Class (Continued)

Application
Class
Bulk
PHB: AF1
Best-Effort
PHB: Default
Example Applications
Database syncs, networkbased backups, Lotus
Notes, Microsoft Outlook,
e-mail download (SMTP,
POP3, IMAP, Exchange),
video content distribution,
large FTP le transfers.
Application/Traffic
Properties
Packet/Message
Sizes
Long le transfers.
Average message
size 64 KB or
greater.
Always invokes TCP

congestion management.
All noncritical trafc,

HTTP web browsing, other
miscellaneous trafc.
DLSw+ Considerations
Some enterprises support legacy IBM equipment that requires data-link switching plus
(DLSw+) to operate across an enterprise environment.
In such cases, it is important to recognize that DLSw+ trafc, by default, is marked to IP
Precedence 5 (DSCP CS5). This default marking could interfere with VoIP provisioning
because both DSCP EF and DSCP CS5 share the same IP Precedence, 802.1Q/p CoS, and
MPLS EXP value (of 5). Therefore, it is recommended to re-mark DLSw+ trafc away
from this default value of IP Precedence 5.
Unfortunately, at the time of writing, Cisco IOS does not support marking DSCP values
within the DLSw+ peering statements; it supports only the marking of type of service (ToS)
values using the dlsw tos map command.
NOTE
To explain this behavior from a historical perspective, when DLSw+ was developed, ToS
was a term loosely used to refer to the rst 3 bits (the IP Precedence bits) of the IP ToS byte,
as dened in RFC 791. For many years, these were typically the only bits of the ToS byte
in use (the others almost always were set to 0), so it seemed to make sense at the time.
However, with the development of newer standards dening the use of the rst 6 bits of the
IP ToS byte for DiffServ markings (RFC 2474) and the last 2 bits for IP explicit congestion
notication (RFC 3168), using the term ToS to refer to only the rst 3 bits (the IP
Precedence bits) of the IP ToS byte has become increasingly inaccurate.
48
However, marking DLSw+ to an IP Precedence/class selector value could interfere with

other QoS Baseline recommended markings. For example, if DLSw+ is marked to IPP 1,
it would be treated as Scavenger trafc; if it is marked to IPP 2, it could interfere with
Network-Management trafc; if it is marked to IPP 3, it could interfere with Call-Signaling;
if it is marked to IPP 4, it would be treated as Streaming-Video trafc; and if it is marked
to IPP 6 or 7, it could interfere with routing or Network Control trafc.
Therefore, a two-step workaround is recommended for marking DLSw+ trafc:
Step 1 Disable native DLSw+ ToS markings with the dlsw tos disable
command.
Step 2 Identify DLSw+ trafc either with access lists (matching the DLSw+
TCP ports 1981 to 1983 or 2065) or with the match protocol dlsw
command.
When DLSw+ trafc is identied, it can be marked as either Transactional (AF21) or
Mission-Critical Data (DSCP 25), depending on the organizations preference.
QoS Requirements of the Control Plane

Unless the network is up, QoS is irrelevant. Therefore, it is critical to provision QoS for
control-plane trafc, which includes IP routing trafc and network management.
IP Routing
When addressing the QoS needs of IP routing trafc, the following guidelines are
recommended:
IP routing trafc should be marked to DSCP CS6; this is default behavior on Cisco
IOS platforms.
Interior gateway protocols usually adequately are protected with the Cisco IOS
internal PAK_priority mechanism. Exterior gateway protocols, such as BGP, are
recommended to have an explicit class for IP routing with a minimal bandwidth
guarantee.
Cisco IOS automatically marks IP routing trafc to DSCP CS6.

By default, Cisco IOS Software (in accordance with RFC 791 and RFC 2474) marks
Interior Gateway Protocol (IGP) trafc (such as Routing Information Protocol [RIP and
RIPv2], Open Shortest Path First [OSPF], and Enhanced Interior Gateway Routing
Protocol [EIGRP]) to DSCP CS6. However, Cisco IOS Software also has an internal
mechanism for granting priority to important control datagrams as they are processed
through the router. This mechanism is called PAK_priority.
Scavenger Class
49
As datagrams are processed though the router and down to the interfaces, they internally
are encapsulated with a small packet header, referred to as the PAKTYPE structure. Within
the elds of this internal header is a PAK_priority ag that indicates the relative importance
of control packets to the routers internal processing systems. PAK_priority designation is
a critical internal Cisco IOS Software operation and, as such, is not administratively
congurable in any way.
It is important to note that although exterior gateway protocol (EGP) trafc, such as Border
Gateway Protocol (BGP) trafc, is marked by default to DSCP CS6, it does not receive such
PAK_priority preferential treatment and might need to be protected explicitly to maintain
peering sessions.
NOTE
Additional information on PAK_priority can be found at http://www.cisco.com/warp/

public/105/rtgupdates.html.
Network-Management
When addressing the QoS needs of Network-Management trafc, the following guidelines
are recommended:
Network-Management trafc should be marked to DSCP CS2.

Network-Management applications should be protected explicitly with a minimal
bandwidth guarantee.
Network-Management trafc is important in performing trend and capacity analyses and

troubleshooting. Therefore, a separate minimal bandwidth queue can be provisioned for
Network-Management trafc, which could include SNMP, NTP, Syslog, and NFS and other
management applications.
Scavenger Class
When addressing the QoS treatment of Scavenger trafc, the following guidelines are
recommended:
Scavenger trafc should be marked to DSCP CS1.

Scavenger trafc should be assigned the lowest congurable queuing service; for
instance, in Cisco IOS, this means assigning a CBWFQ of 1 percent to Scavenger.
The Scavenger class is intended to provide deferential services, or less-than best-effort

services, to certain applications. Applications assigned to this class have little or no contribution to the organizational objectives of the enterprise and are typically entertainment
oriented in nature. These include peer-to-peer media-sharing applications (KaZaa, Morpheus,
50
Groekster, Napster, iMesh, and so on), gaming applications (Doom, Quake, Unreal Tournament, and so on), and any entertainment video applications.
Assigning Scavenger trafc to minimal bandwidth queue forces it to be squelched to
virtually nothing during periods of congestion, but it allows it to be available if bandwidth
is not being used for business purposes, such as might occur during off-peak hours.
The Scavenger class is a critical component to the DoS and worm mitigation strategy,
discussed next.
DoS and Worm Mitigation Strategy Through Scavenger

Class QoS
Worms are nothing new; they have been around in some form since the beginning of the
Internet and steadily have been increasing in complexity, as shown in Figure 2-6.
Business Security Threat Evolution
Global
Impact
Scope of Damage
Figure 2-6
Next Gen
Regional
Networks
Multiple
Networks
3rd Gen
2nd Gen
Individual
Networks
Individual
Computer
1st Gen
Boot Viruses
1980s
Macro Viruses,
Trojans, E-Mail,
Single-Server
DoS, Limited
Targeted
Hacking
1990s
Multiserver
DoS, DDoS,
Blended Threat
(Worm+ Virus+
Trojan), Turbo
Worms,
Widespread
System Hacking
Today
Infrastructure
Hacking, Flash
Threats,
Massive WormDriven DDoS,
Negative
Payload
Viruses,
Worms, and
Trojans
Future
Sophistication of Threats
Particularly since 2002, there has been an exponential increase not only in the frequency of
DoS and worm attacks, but also in their relative sophistication and scope of damage. For
example, more than 994 new Win32 viruses and worms were documented in the rst half
of 2003, more than double the 445 documented in the rst half of 2002. Some of these more
recent worms are shown in Figure 2-7.
DoS and Worm Mitigation Strategy Through Scavenger Class QoS
Figure 2-7
51
Recent Internet Worms
Code Red/
S sadmind/IIS
Code Red
NIMDA
Apache/
Apach e/
mod_ssl
MS-SQL
Slammer
W32/
Blaster
W32/Sobig
W32/
MyDoom
MyDoo m
W32/Bagel
Sasser
May 01
Sep 01
Jul 02
Jan 03
Aug 03
Jan 04
April 04
DoS or worm attacks can be categorized into two main classes:
Spoong attacksThe attacker pretends to provide a legitimate service but provides

false information (if any) to the requester.
Flooding attacksThe attacker exponentially generates and propagates trafc until

service resources (servers or network infrastructure) are overwhelmed.
Spoong attacks best are addressed by authentication and encryption technologies;

ooding attacks, on the other hand, can be mitigated using QoS technologies.
The majority of ooding attacks target PCs and servers, which, when infected, target other
PCs and servers, thus multiplying trafc ows. Network devices themselves are not usually
the direct targets of attacks. But the rapidly multiplying volumes of trafc ows eventually
drown the CPU and hardware resources of routers and switches in their paths, causing
denial of service to legitimate trafc ows. The end result is that network devices become
indirect victims of the attack. This is illustrated in Figure 2-8.
A reactive approach to mitigating such attacks is to reverse-engineer the worm and set up
intrusion-detection mechanisms or ACLs to limit its propagation. However, the increased
sophistication and complexity of todays worms make them harder to identify from legitimate trafc ows. This exacerbates the nite time lag between when a worm begins to propagate and when the following occurs:
Sufcient analysis has been performed to understand how the worm operates and
what its network characteristics are.
An appropriate patch, plug, or ACL is disseminated to network devices that might be

in the path of the worm. This task might be hampered by the attack itself because
network devices might become unreachable for administration during the attacks.
52
Figure 2-8
Impact of an Internet WormDirect and Collateral Damage
Systems
Under
Attack
Infected
Source
Si
Core
Distribution
Access
End Systems
Overloaded
High CPU
Applications Impacted
Network Links
Overloaded
Routers
Overloaded
High Packet Loss

Mission-Critical
Applications Impacted
High CPU
Instability
Loss of Management
These time lags might not seem long in absolute terms, such as in minutes, but the relative
window of opportunity for damage is huge. For example, in 2003, the number of hosts
infected with the Slammer worm (a Sapphire worm variant) doubled every 8.5 seconds
on average, infecting more than 75,000 hosts in just 11 minutes and performing scans of
55 million more hosts within the same time period.
NOTE
Interestingly, a 2002 CSI/FBI report stated that the majority of network attacks occur from
within an organization, typically by disgruntled employees.
A proactive approach to mitigating DoS and worm ooding attacks within enterprise networks is to respond immediately to out-of-prole network behavior indicative of a DoS
or worm attack via campus Access-Layer policers. Such policers can meter trafc rates
received from endpoint devices and, when these exceed specied watermarks (at which
point they no longer are considered normal ows), can mark down excess trafc to the
Scavenger class marking (DSCP CS1).
DoS and Worm Mitigation Strategy Through Scavenger Class QoS
53
In this respect, the policers would be fairly dumb. They would not be matching specic
network characteristics of specic types of attacks, but they simply would be metering trafc volumes and responding to abnormally high volumes as close to the source as possible.
The simplicity of this approach negates the need for the policers to be programmed with
knowledge of the specic details of how the attack is being generated or propagated. It is
precisely this dumbness of such Access-Layer policers that allows them to maintain relevancy as worms mutate and become more complex: The policers dont care how the trafc
was generated or what it looks like; all they care about is how much trafc is being put onto
the wire. Therefore, they continue to police even advanced worms that continually change
the tactics of how trafc is being generated.
For example, in most enterprises, it is quite abnormal (within a 95 percent statistical
condence interval) for PCs to generate sustained trafc in excess of 5 percent of their
links capacity. In the case of a Fast Ethernet switch port, this means that it would be
unusual in most organizations for an end users PC to generate more than 5 Mbps of uplink
trafc on a sustained basis.
NOTE
It is important to recognize that this value ( 5 percent) for normal access-edge utilization
by endpoints is just an example value. This value would likely vary from industry vertical
to vertical, and from enterprise to enterprise.
It is important to recognize that what is being proposed is not to police all trafc to 5 Mbps
and automatically drop the excess. If that were the case, there would not be much reason
for deploying Fast Ethernet or Gigabit Ethernet switch ports to endpoint devices because
even 10BASE-T Ethernet switch ports would have more uplink capacity than a 5 Mbps
policer-enforced limit. Furthermore, such an approach supremely would penalize legitimate trafc that exceeded 5 Mbps on a Fast Ethernet switch port.
A less draconian approach is to couple Access-Layer policers with hardware and software
(campus, WAN, and VPN) queuing policies, with both sets of policies provisioning for a
less-than best-effort Scavenger class.
This would work by having Access-Layer policers mark down out-of-prole trafc to
DSCP CS1 (Scavenger) and then have all congestion-management policies (whether in
Catalyst hardware or in Cisco IOS Software) provision a less-than best-effort service for
any trafc marked to DSCP CS1.
Lets examine how this might work, for both legitimate trafc exceeding the Access-Layer
policers watermark and illegitimate excess trafc (the result of a DoS or worm attack).
In the former case, imagine that the PC generates more than 5 Mbps of trafc, perhaps
because of a large le transfer or backup. Because there is generally abundant capacity
within the campus to carry the trafc, congestion (under normal operating conditions) is
54
rarely, if ever, experienced. Typically, the uplinks to the distribution and core layers of
the campus network are Gigabit Ethernet, which requires 1000 Mbps of trafc from the
Access-Layer switch to create congestion. If the trafc was destined to the far side of a
WAN or VPN link (which are rarely more than 5 Mbps in speed), dropping would occur
even without the Access-Layer policer, simply because of the campus/WAN speed mismatch and resulting bottleneck. TCPs sliding-windows mechanism eventually would nd
an optimal speed (less than 5 Mbps) for the le transfer.
To make a long story short, Access-Layer policers that mark down out-of-prole trafc to
Scavenger (CS1) would not affect legitimate trafc, aside from the obvious re-marking. No
reordering or dropping would occur on such ows as a result of these policers (that would
not have occurred anyway).
In the latter case, the effect of Access-Layer policers on trafc caused by DoS or worm
attacks is quite different. As hosts become infected and trafc volumes multiply, congestion
might be experienced even within the campus. If just 11 end-user PCs on a single switch
begin spawning worm ows to their maximum Fast Ethernet link capacities, the GE uplink
from the Access-Layer switch to the Distribution-Layer switch will congest and queuing or
reordering will engage. At such a point, VoIP and critical data applications, and even BestEffort applications, would gain priority over worm-generated trafc (and Scavenger trafc
would be dropped the most aggressively); network devices would remain accessible for
administration of patches, plugs, and ACLs required to fully neutralize the specic attack.
WAN links also would be protected: VoIP, critical data, and even best-effort ows would
continue to receive priority over any trafc marked down to Scavenger/CS1. This is a huge
advantage because generally WAN links are the rst to be overwhelmed by DoS and worm
attacks. The bottom line is that Access-Layer policers signicantly mitigate network trafc
generated by DoS or worm attacks.
It is important to recognize the distinction between mitigating an attack and preventing it
entirely: The strategy being presented does not guarantee that no denial of service or worm
attacks ever will happen, but it can reduce the risk and impact that such attacks could have
on the network infrastructure.
Principles of QoS Design

The richness of the Cisco QoS toolset allows for a myriad of QoS design and deployment
options. However, a few succinct design principles can help simplify strategic QoS designs
and lead to an expedited, cohesive, and holistic end-to-end deployment. Some of these
design principles are summarized here; others, which are LAN-, WAN- or VPN-specic,
are covered in detail in their respective design chapters.
55
General QoS Design Principles

A good place to begin is to decide which comes rst: the cart or the horse. The horse, in
this context, serves to pull the cart and is the enabler for this objective. Similarly, QoS
technologies are simply the enablers to organizational objectives. Therefore, the way to
begin a QoS deployment is not by glossing over the QoS toolset and picking la carte tools
to deploy. In other words, do not enable QoS features simply because they exist. Instead,
start from a high level and clearly dene the organizational objectives.
Some questions for high-level consideration include the following:
Is the objective to enable VoIP only?

Is video also required? If so, what type(s) of video: interactive or streaming?
Are some applications considered mission critical? If so, what are they?
Does the organization want to squelch certain types of trafc? If so, what are they?
All trafc classes specied in the QoS Baseline model except onethe Locally-Dened,
Mission-Critical Data application classare determined by objective networking characteristics. These applications, a subset of the Transactional Data class, are selected for a dedicated, preferential class of service because of their signicant impact on the organizations
main business objectives.
This is usually a highly subjective evaluation that can excite considerable controversy and
dispute. An important principle to remember when assigning applications to the MissionCritical Data class is that as few applications as possible should be assigned to the LocallyDened Mission-Critical class.
If too many applications are assigned to it, the Mission-Critical Data class will dampen, and
possibly even negate, the value of having a separate class (from Transactional Data). For
example, if 10 applications are assigned as Transactional Data (because of their interactive,
foreground networking characteristics) and all 10 are determined to be classied as
Mission-Critical Data, the whole point of a separate class for these applications becomes
moot. However, if only one or two of the Transactional Data applications are assigned to
the Mission-Critical Data class, the class will prove highly effective.
Related to this point, it is recommended always to seek executive endorsement of the QoS
objectives before design and deployment. By its very nature, QoS is a system of managed
unfairness and, as such, almost always creates political and organizational repercussions
when implemented. To minimize the effects of such nontechnical obstacles to deployment,
which could prevent the QoS implementation altogether, it is recommended to address
these political and organizational issues as early as possible and to solicit executive
endorsement whenever possible.
As stated previously, it is not mandated that enterprises deploy all 11 classes of the QoS
Baseline model; this model is designed to be a forward-looking guide for consideration of
the many classes of trafc that have unique QoS requirements. Being aware of this model
56
can help bring about a smooth expansion of QoS policies to support additional applications
as future requirements arise. However, at the time of QoS deployment, the organization
needs to clearly dene how many classes of trafc are required to meet the organizational
objectives.
This consideration should be tempered with the consideration of how many classes of
applications the networking administration team feels comfortable with deploying and
supporting. Platform-specic constraints or service-provider constraints also might come
into play when arriving at the number of classes of service. At this point, it also would be
good to consider a migration strategy to allow the number of classes to be expanded
smoothly as future needs arise, as illustrated in Figure 2-9.
Figure 2-9
Example Strategy for Expanding the Number of Classes of Service over Time
4/5 Class Model
8 Class Model
QoS Baseline Model
Voice
Voice
Real-Time
Call-Signaling
Interactive-Video
Video
Streaming-Video
Call-Signaling
Call-Signaling
IP Routing
Network Control
Critical Data
Critical Data
Network-Management
Transactional Data
Bulk Data
Bulk Data
Best-Effort
Best-Effort
Best-Effort
Scavenger
Scavenger
Scavenger
Time
When the number of classes of service has been determined, the details of the required
marking, policing, and queuing policies can be addressed. When deciding where to enable
such policies, keep in mind that QoS policies always should be performed in hardware
instead of software whenever a choice exists.
Cisco IOS routers perform QoS in software, which places incremental loads on the CPU
(depending on the complexity and functionality of the policy). Cisco Catalyst switches, on
the other hand, perform QoS in dedicated hardware ASICS and, as such, do not tax their
main CPUs to administer QoS policies. This allows complex policies to be applied at line
rates at even 1-Gbps or 10-Gigabit speeds.
57
Classication and Marking Principles

When it comes to classifying and marking trafc, an unofcial Differentiated Services
design principle is to classify and mark applications as close to their sources as technically
and administratively feasible. This principle promotes end-to-end Differentiated Services
and per-hop behaviors (PHBs). Sometimes endpoints can be trusted to set CoS and DSCP
markings correctly, but, in most cases, it is not a good idea to trust markings that users
can set on their PCs (or other similar devices). This is because users easily could abuse
provisioned QoS policies if permitted to mark their own trafc. For example, if DSCP EF
receives priority services throughout the enterprise, a user easily could congure the PC to
mark all trafc to DSCP EF right on the NIC, thus hijacking network-priority queues to
service that users non-real-time trafc. Such abuse easily could ruin the service quality of
real-time applications (such as VoIP) throughout the enterprise. For this reason, the clause
as close as . . . administratively feasible is included in the design principle.
Following this rule, it further is recommended to use DSCP markings whenever possible
because these are end to end, more granular, and more extensible than Layer 2 markings.
Layer 2 markings are lost when media changes (such as at a LAN-to-WAN or VPN edge).
An additional constraint to Layer 2 marking is that there is less marking granularity; for
example, 802.1Q/p CoS supports only 3 bits (values 0 through 7), as does MPLS EXP.
Therefore, only (up to) eight classes of trafc can be supported at Layer 2, and interclass
relative priority (such as RFC 2597 assured-forwarding class markdown) is not supported.
On the other hand, Layer 3 DSCP markings allow for up to 64 classes of trafc, which is
more than enough for most enterprise requirements for the foreseeable future.
Because the line between enterprises and service providers is blurring and the need for
interoperability and complementary QoS markings is critical, it is recommended to follow
standards-based DSCP PHB markings to ensure interoperability and future expansion.
The QoS Baseline marking recommendations are standards based, making it easier for
enterprises adopting these markings to interface with service provider classes of service.
Network mergers are also easier to manage when standards-based DSCP markings are
used, whether these mergers are the result of acquisitions, partnerships, or strategic
alliances.
Policing and Markdown Principles

There is little sense in forwarding unwanted trafc only to police and drop it at a subsequent
node. This is especially the case when the unwanted trafc is the result of DoS or worm
attacks. The overwhelming volumes of trafc that such attacks can create readily can drive
network device processors to their maximum levels, causing network outages. Therefore, it
is recommended to police trafc ows as close to their sources as possible. This principle
applies to legitimate ows also because DoS and worm-generated trafc might be masquerading under legitimate, well-known TCP and UDP ports, causing extreme amounts of
58
trafc to be poured onto the network infrastructure. Such excesses should be monitored at
the source and marked down appropriately.
Whenever supported, markdown should be done according to standards-based rules, such
as RFC 2597 (Assured Forwarding PHB Group). In other words, whenever supported,
trafc marked to AFx1 should be marked down to AFx2 or AFx3. For example, in the case
of a single-rate policer, excess trafc originally marked AF11 should be marked down to
AF12. In the case of a dual-rate policer (as dened in RFC 2698), excess trafc originally
marked AF11 should be marked down to AF12, and violating trafc should be marked
down further to AF13. Following such markdowns, congestion-management policies, such
as DSCP-based WRED, should be congured to drop AFx3 more aggressively than AFx2,
which, in turn, is dropped more aggressively than AFx1.
However, at the time of writing, Cisco Catalyst switches do not perform DSCP-based
WRED, so this standards-based strategy cannot be implemented fully. As an alternative
workaround, single-rate policers can be congured to mark down excess trafc to DSCP
CS1 (Scavenger); dual-rate policers can be congured to mark down excess trafc to AFx2,
while marking down violating trafc to DSCP CS1. Such workarounds yield an overall
similar effect as the standards-based policing model. However, when DSCP-based WRED
is supported on all routing and switching platforms, it would be more standards compliant
to mark down assured-forwarding classes by RFC 2597 rules.
Queuing and Dropping Principles

Critical applications, such as VoIP, require service guarantees regardless of network
conditions. The only way to provide service guarantees is to enable queuing at any node
that has the potential for congestionregardless of how rarely, in fact, this might occur.
This principle applies not only to campus-to-WAN or VPN edges, where speed mismatches
are most pronounced, but also to campus interlayer links (where oversubscription ratios
create the potential for congestion). There is simply no other way to guarantee service
levels than to enable queuing wherever a speed mismatch exists.
When provisioning queuing, some best-practice rules of thumb also apply. For example, as
discussed previously, the Best-Effort class is the default class for all data trafc. Only if an
application has been selected for preferential or deferential treatment is it removed from the
default class. Because many enterprises have several hundred, if not thousands of, data
applications running over their networks, adequate bandwidth must be provisioned for this
class as a whole to handle the sheer volume of applications that default to it. Therefore, it
is recommended that at least 25 percent of a links bandwidth be reserved for the default
Best-Effort class.
Another class of trafc that requires special consideration when provisioning queuing is
the Real-Time or Strict-Priority class (which corresponds to RFC 3246, An Expedited
Forwarding Per-Hop Behavior). The amount of bandwidth assigned to the Real-Time
59
queuing class is variable. However, if too much trafc is assigned for strict-priority
queuing, the overall effect is a dampening of QoS functionality for non-real-time
applications.
The goal of convergence cannot be overemphasized: to enable voice, video, and data to
coexist transparently on a single network. When real-time applications (such as Voice or
Interactive-Video) dominate a link (especially a WAN/VPN link), data applications will
uctuate signicantly in their response times, destroying the transparency of the
converged network.
Cisco Technical Marketing testing has shown a signicant decrease in data application
response times when real-time trafc exceeds one-third of a links bandwidth capacity.
Extensive testing and customer deployments have shown that a general best queuing
practice is to limit the amount of strict-priority queuing to 33 percent of a links capacity.
This strict-priority queuing rule is a conservative and safe design ratio for merging realtime applications with data applications.
Cisco IOS Software allows the abstraction (and, thus, conguration) of multiple (strictpriority) low-latency queues. In such a multiple-LLQ context, this design principle applies
to the sum of all LLQs: They should be within one-third of a links capacity.
NOTE
This strict-priority queuing rule (limit to 33 percent) is simply a best-practice design

recommendation; it is not a mandate. In some cases, specic business objectives cannot be
met while holding to this recommendation. In such cases, enterprises must provision
according to their detailed requirements and constraints. However, it is important to
recognize the trade-offs involved with overprovisioning strict-priority trafc with respect
to the negative performance impact on response times in non-real-time applications.
Whenever a Scavenger queuing class is enabled, it should be assigned a minimal amount

of bandwidth. On some platforms, queuing distinctions between Bulk Data and Scavenger
class trafc ows cannot be made because queuing assignments are determined by CoS
values, and these applications share the same CoS value of 1. In such cases, the Scavenger/
Bulk Data queuing class can be assigned a bandwidth percentage of 5. If Scavenger and
Bulk trafc can be assigned uniquely to different queues, the Scavenger queue should be
assigned a bandwidth percentage of 1.
The Real-Time, Best-Effort, and Scavenger classes queuing best-practice principles are
60
Figure 2-10 Real-Time, Best-Effort, and Scavenger Queuing Rules
Best-Effort
25%
Real-Time
33%
Scavenger/Bulk
5%
Critical Data
Some platforms support different queuing structures than others. To ensure consistent
PHBs, congure consistent queuing policies according to platform capabilities.
For example, on a platform that supports only four queues with CoS-based admission (such
as a Catalyst switch), a basic queuing policy could be as follows:
Real-Time ( 33 percent)
Critical Data
Best-Effort ( 25 percent)
Scavenger/Bulk(< 5 percent)
However, on a platform that supports a full QoS Baseline queuing model, the queuing
policies can be expanded, yet in such a way that they provide consistent servicing to RealTime, Best-Effort, and Scavenger class trafc. For example, on a platform that supports 11
queues with DSCP-based admission (such as a Cisco IOS router), an advanced queuing
policy could be as follows:
Voice ( 18 percent)
Interactive-Video ( 15 percent)
Internetwork Control
Call-Signaling
Transactional Data
Network-Management
Streaming-Video Control
61
Best-Effort ( 25 percent)
Bulk Data (4 percent)
Scavenger (1 percent)
Figure 2-11 illustrates the interrelationship between these compatible queuing models.
Figure 2-11 Compatible 4-Class and 11-Class Queuing Models Following Real-Time, Best-Effort, and Scavenger
Class Queuing Rules
Voice 18%
Best-Effort
25%
Scavenger
1%
Scavenger/
Bulk Data 5%
Best-Effort
25%
Bulk Data 4%
Real-Time
33%
Interactive-Video
15%
Critical Data
Streaming-Video
Internetwork
Control
Network-Management
Call-Signaling
Transactional Data Mission-Critical Data
In this manner, trafc will receive compatible queuing at each node, regardless of platform
capabilitieswhich is the overall objective of DiffServ per-hop behavior denitions.
Whenever supported, it is recommended to enable WRED (preferably DSCP-based WRED)
on all TCP ows. In this manner, WRED congestion avoidance will prevent TCP global
synchronization and will increase overall throughput and link efciency. Enabling WRED
on UDP ows is optional.
DoS and Worm Mitigation Principles

Whenever part of the organizations objectives is to mitigate DoS and worm attacks through
Scavenger-class QoS, the following best practices apply.
First, the network administrators need to prole applications to determine what constitutes
normal versus abnormal ows, within a 95 percent condence interval. Thresholds
62
differentiating normal and abnormal ows vary from enterprise to enterprise and from
application to application. Caution must be extended not to overscrutinize trafc behavior
because this could be time and resource exhaustive and easily could change from one day
to the next. Remember, the presented Scavenger-class strategy will not apply a penalty to
legitimate trafc ows that exceed thresholds (aside from re-marking); only sustained,
abnormal streams generated simultaneously by multiple hosts (highly indicative of DoS
and worm attacks) are subject to aggressive dropping, and only after legitimate trafc has
been serviced.
To contain such abnormal ows, it is recommended to deploy campus Access-Edge policers
to re-mark abnormal trafc to Scavenger (DSCP CS1). Additionally, whenever Catalyst
6500s with Supervisor 720s are deployed in the distribution layer, it is recommended to
deploy a second line of policing defense, at the distribution layer via per-user microow
policing.
To complement these re-marking policies, it is necessary to enforce end-to-end Scavengerclass queuing policies, where ows marked as Scavenger will receive a less-than best-effort
service whenever congestion occurs.
It is important to note that even when Scavenger-class QoS has been deployed end to end,
this strategy only mitigates DoS and worm attacks and does not prevent them or remove
them entirely. Therefore, it is critical to overlay security, rewall, intrusion detection, and
identity systems, along with Cisco Guard and Cisco Security Agent solutions, on top of the
QoS-enabled network infrastructure.
Deployment Principles
After the QoS designs have been nalized, it is vital that the networking team thoroughly
understand the QoS features and syntax before enabling features on production networks.
Such knowledge is critical for both deployment and troubleshooting QoS-related issues.
Furthermore, it is a general best practice to schedule proof-of-concept (PoC) tests to verify
that the hardware and software platforms in production support the required QoS features
in combination with all the other features that they currently are running. Remember, in
theory, theory and practice are the same. In other words, there is no substitute for testing.
When testing has validated the designs, it is recommended to schedule network downtime
to deploy QoS features. Although QoS is required end to end, it does not have to be
deployed end to end at a single instance. A pilot network segment can be selected for an
initial deployment, and, pending observation, the deployment can be expanded in stages to
encompass the entire enterprise. A rollback strategy always is recommended, to address
unexpected issues that arise from the QoS deployment.
Summary
63
Summary
This chapter began by reviewing the QoS requirements of voice, video, and data
applications.
Voice requires 150-ms one-way, end-to-end (mouth-to-ear) delay; 30 ms of one-way jitter;
and no more than 1 percent packet loss. Voice should receive strict-priority servicing, and
the amount of priority bandwidth assigned for it should take into account the VoIP codec;
the packetization rate; IP, UDP, and RTP headers (compressed or not); and Layer 2
overhead. Additionally, provisioning QoS for IP Telephony requires that a minimal amount
of guaranteed bandwidth be allocated to Call-Signaling trafc.
Video comes in two avors: Interactive-Video and Streaming-Video. Interactive-Video has
the same service-level requirements as VoIP because embedded within the video stream is
a voice call. Streaming-Video has much laxer requirements because of a high amount of
buffering that has been built into the applications.
Control plane requirements, such as provisioning moderate bandwidth guarantees for IP
routing protocols and network-management protocols, should not be overlooked.
Data comes all shapes and sizes but generally can be classied into four main classes: BestEffort (the default class), Bulk Data (noninteractive background ows), Transactional/
Interactive (interactive, foreground ows), and Mission-Critical Data. Mission-Critical
Data applications are locally dened, meaning that each organization must determine the
select few Transactional Data applications that contribute the most signicantly to its
overall business objectives.
A less-than best-effort Scavenger class of trafc was introduced, and a strategy for using
this class for DoS and worm mitigation was presented. Specically, ows can be monitored
at the campus Access-Edge, and out-of-prole ows can be marked down to the Scavenger
marking (of DSCP CS1). To complement these policers, queues providing a less-than besteffort Scavenger service during periods of congestion are deployed in the LAN, WAN,
and VPN.
The chapter concluded with a set of general best-practice principles relating to QoS
planning, classication, marking, policing, queuing, and deployment. These best practices
include the following:
Clearly dening the organizations business objectives to be addressed by QoS
Performing QoS functions in (Catalyst switch) hardware instead of (Cisco IOS router)
software, whenever possible
Classifying trafc as close to the source as administratively feasible, preferably at

Layer 3 with standards-based DSCP markings
Selecting an appropriate number of service classes to meet these business objectives

Soliciting executive endorsement, whenever possible, of the trafc classications,
especially when determining any mission-critical applications
64
Policing trafc as close to the source as possible, following standards-based rules

(such as RFC 2597, Assured Forwarding Markdown), whenever possible
Provisioning at least one-quarter of a link to service Best-Effort trafc
Provisioning a less-than best-effort Scavenger queue, which should be assigned as

low of a bandwidth allocation as possible
Understanding and thoroughly testing desired QoS features in conjunction with

features already enabled on the production network
Deploying end-to-end QoS in stages during scheduled network downtime, with a

recommended rollback strategy
Provisioning no more than one-third of a link to service real-time and strict-priority

trafc
Further Reading
Standards:
RFC 791, Internet Protocol Specication: http://www.ietf.org/rfc/rfc791.
RFC 2597, Assured Forwarding PHB Group: http://www.ietf.org/rfc/rfc2597.
RFC 3246, An Expedited Forwarding PHB (Per-Hop Behavior): http://

www.ietf.org/rfc/rfc3246.
IPv6 Headers: http://www.ietf.org/rfc/rfc2474.
RFC 2697, A Single Rate Three Color Marker: http://www.ietf.org/rfc/rfc2697.
RFC 2698, A Two Rate Three Color Marker: http://www.ietf.org/rfc/rfc2698.
RFC 3168, The Addition of Explicit Congestion Notication (ECN) to IP: http://
www.ietf.org/rfc/rfc3168.
Cisco documentation:
Cisco IOS QoS Conguration Guide, Cisco IOS version 12.3: http://www.cisco.com/
univercd/cc/td/doc/product/software/ios123/123cgcr/qos_vcg.htm.
Cisco IOS Conguration GuideConguring Data Link Switching Plus, Cisco IOS
version 12.3: http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/
122cgcr/bm_c/bcfpart2/bcfdlsw.htm.
Understanding how routing updates and Layer 2 control packets are queued on an
interface with a QoS service policy (PAK_priority): http://www.cisco.com/warp/
public/105/rtgupdates.html.
PA R T
II
QoS Toolset
Part II of this book provides context for the design chapters that follow by presenting a brief overview
of Cisco QoS tools. Particular emphasis is placed on tool-specic idiosyncrasies and tool-interaction
caveats (many of which are not included in IOS documentation). A basic understanding and familiarity of
these QoS tools is assumed.
Chapter 3
Chapter 4
Chapter 5
Chapter 6
Chapter 7
Link-Specic Tools
Chapter 8
Chapter 9
Chapter 10
Catalyst QoS Tools
Chapter 11
WLAN QoS Tools
This chapter includes the following topics:
Classication and marking
Packet marking in different technologies, such as IP, MPLS, ATM, Frame Relay, and
Ethernet
Class-based classication and marking techniques and other mechanisms to achieve

these results
Discussion of Layer 2 and Layer 3 marking elds and how these translate to
each other
CHAPTER

The rst step in dening a Quality-of-Service (QoS) policy is to identify the trafc that is to
be treated differently (either preferentially or differentially). This is accomplished through
classication and marking.
Although the terms classication and marking often are used interchangeably, the terms
represent distinct and different actions that work together but also can be used
independently.
Classication tools sort packets into different trafc types, to which different policies
then can be applied. The classication of packets normally occurs at each node in the
network but is not required to be done everywhere. Classication of packets can
happen without marking.
Marking (or re-marking) typically establishes a trust boundary on which scheduling

tools later depend. The network edge where markings are accepted (or rejected) is
referred to as the trust-boundary. Marking also can be used in other locations in the
network, as necessary, and is not always used solely for purposes of classication.
As with the general terms classication and marking, there is a difference in the action that
the actual tools, named classiers and markers, take on trafc.
ClassiersInspect one or more elds in a packet to identify the type of trafc that
the packet is carrying. After being identied, the trafc is directed to the applicable
policy-enforcement mechanism for that trafc type, where it receives predened
treatment (either preferential or deferential). Such treatment can include marking
and re-marking, queuing, policing, shaping, or any combination of these (and other)
actions.
MarkersWrite a eld within the packet, frame, cell, or label to preserve the
classication decision that was reached at the trust boundary. By marking trafc at
the trust boundary edge, subsequent nodes do not have to perform the same in-depth
classication and analysis to determine how to treat the packet.
70
Chapter 3: Classication and Marking Tools
Classication Tools
Classication tools examine any of the following criteria to identify a ow and assign it for
preferential or deferential treatment:
Layer 1 (L1) parametersPhysical interface, subinterface, PVC, or port
Layer 3 (L3) parametersIP Precedence, DiffServ code point (DSCP),

source/destination IP address
Layer 4 (L4) parametersTCP or User Datagram Protocol (UDP) ports
Layer 2 (L2) parametersMAC address, 802.1Q/p class of service (CoS) bits,

VLAN identication, experimental bits (MPLS EXP), ATM cell loss priority (CLP),
and Frame Relay discard eligible (DE) bits
Layer 7 (L7) parametersApplication signatures and uniform resource locators

(URLs) in packet headers or payload
Figure 3-1 shows the progressive depth at which a frame or packet may be examined to
make a classication decision. It is not shown to scale because of space limitations.
Figure 3-1
Frame/Packet Classication Fields
TCP/UDP
Segment
IP Packet
ToS/
DSCP
NOTE
Source
IP
Dest
IP
Src
Port
Dst
Port
Data Payload
NBAR PDLM
Figure 3-1 is intended to represent only the comparisons of data-link, network, transport,
and application layer QoS ltering criteria and, therefore, many elds have been omitted
and the diagram is not to scale.
Only after trafc is positively identied can policies be applied to it. Therefore, bestpractice design recommendations are to identify and mark trafc (with DSCP values) as
close to the source of the trafc as possible, typically in the wiring closet or within the
trusted devices (such as IP phones) themselves. If markings and trusts are set correctly, the
intermediate hops do not have to repeat the same in-depth classication. Instead, they can
administer QoS policies (such as scheduling) based on the previously set markings, which
appear close to the beginning of the frame or packet.
Classication Tools
71
Modular QoS Command-Line Interface Class Maps

The principle tool for QoS classication within Cisco IOS today is modular QoS CLI
(MQC)based class maps. Class maps identify trafc ows using a wide array of ltering
criteria, which are individually dened by match statements within the class map. Multiple
match statements can be dened under a single class map. When multiple match statements
are used, the class map can be specied as follows:
match-allA logical AND operand, meaning that all match statements must be true
at the same time for the class map condition to be true
match-anyA logical OR operand, meaning that any of the match statements can
be true for the class map condition to be true
Including match-any or match-all when dening a class map is optional, but it is important to note that if neither is specied, the default behavior is match-all. For example, if
class-map FOO is entered, the Cisco IOS parser actually expands this to class-map matchall FOO within the conguration. Example 3-1 illustrates the matching criteria available
within MQC class-maps.
Example 3-1 match-all as Default Cisco IOS Behavior
Router(config) class-map FOO
Router(config-cmap)#match ?
access-group
Access group
any
Any packets
class-map
Class map
cos
IEEE 802.1Q/ISL class of service/user priority values
destination-address
Destination address
input-interface
Select an input interface to match
ip
IP specific values
mpls
Multi Protocol Label Switching specific values
not
Negate this match result
protocol
Protocol
qos-group
Qos-group
source-address
Source address
Although the sequence in which class maps are dened within the conguration is unimportant,
the sequence of classes within a policy map is important. This is because, as with access list
(ACL) logic, policy maps apply the First-True-Match rule, meaning that the classes examine a
packet until a match is found. When a match is found, the classication process nishes and
no further class maps are checked. If no matches are found, the packet ends up in an implicit
class default, which essentially means everything else.
For example, consider the service policy shown in Example 3-2 that illustrates the classication of two classes of trafc: VOICE for voice trafc and FAX-RELAY for fax trafc.
The sequence of class-map FAX-RELAY and class-map VOICE within the global conguration does not matter to the classication functionality; these can be entered in any
order. The assumption in this example is that both voice trafc and fax-relay trafc are
72
marked to DSCP EF at their respective sources. Therefore, the question is how to treat these
two trafc types differently because they are marked the same.
The policy map shown in Example 3-2 is unusual, although valid, in two respects:
Multiple priority classesBoth voice and fax trafc must be prioritized (as covered
in Chapter 5, Congestion-Management Tools, a later chapter on queuing). Typically,
both these trafc classes would be handled by a single class denition, but in this case,
the desire was to control strictly the bandwidth used by each class of trafc. This
required two different priority class denitions.
Police statements in the priority classesNormally, priority class trafc is not

explicitly policed, as IOS has an implicit policer, which is discussed in additional
detail in Chapter 5, Congestion-Management Tools, to prevent the starvation of
other queues. However, this example shows how a service-level agreement (SLA) can
be strictly enforced so that different classes of trafc cannot exceed the agreed-upon
bandwidth allocation.
Example 3-2 Class Denition Sequence in Policy Map

Router#show run
class-map match-all FAX-RELAY
match ip dscp ef
class-map match-all VOICE
match protocol rtp audio
!
policy-map VOICE-AND-FAX
class VOICE
priority 216
police cir 216000
class FAX-RELAY
priority 64
police cir 64000
class class-default
fair-queue
The policy map VOICE-AND-FAX provides the answer through careful ordering of the
classes within it. First, all packets are checked against the class VOICE, which performs
Network-Based Application Recognition (NBAR) classication to identify whether the
trafc is Real-Time Protocol audio (in other words, voice). Only trafc that fails this
examination is checked against the second class under the policy map (the class FAXRELAY).
The class FAX-RELAY checks whether the packets DSCP value is EF. Because only two
types of trafc can have DSCP values of EF (voice and fax-relay) and voice has already
been ltered out, any remaining trafc that matches these criteria must be fax-relay. Faxrelay trafc then is administratively assigned a slightly different treatment. The details of
the treatment in this example are irrelevant. The emphasis is on how the ordering of the
classes within policy maps can offer more granular classication options because of the
Classication Tools
73
First-True-Match logic that policy maps employ. If the sequence of these two statements
were reversed, the policy would work very differently: No trafc would ever show against
the VOICE class because both voice and fax-relay trafc would be matched on DSCP EF
and would be assigned to the FAX-RELAY class. All other trafc would fall into the
implicit class-default class. Figure 3-2 shows the decision hierarchy for each packet
examined by the policy map VOICE-AND-FAX.
Figure 3-2
Classication Decisions by Policy Map VOICE-AND-FAX

New Packet Arrives
Class VOICE Classifier
NBAR: RTP
Packet?
Yes
Police to 216 kbps

Bandwidth and Transmit
No
Class FAX-RELAY Classifier
DSCP EF?
Yes
Police to 64 kbps
Bandwidth and Transmit
No
Class class-default
It is important to note that class map and policy map names (similar to ACL names) are case
sensitive to the Cisco IOS. Thus, class-map foo is different from class-map Foo, which is
different from class-map FOO. Therefore, it is very important that the class map names
and cases match exactly to the class names called out under policy maps. In this book, such
names are shown in uppercase letters to clearly distinguish them from Cisco IOS commands.
This is entirely an administrative preference.
Network-Based Application Recognition

Although the majority of data applications can be identied using Layer 3 or Layer 4 criteria
(such as discrete IP addresses or well-known TCP/UDP ports), some applications cannot
be identied by such criteria alone. This might be because of legacy limitations, but more
likely it is by deliberate design. For example, peer-to-peer media-sharing applications (such
as KaZaa, Morpheus, and Napster) deliberately negotiate ports dynamically with the objective
of penetrating rewalls.
74
When Layer 3 or 4 parameters are insufcient to positively identify an application, NBAR

is a viable alternative solution.
NBAR is the most sophisticated classier in the Cisco IOS tool suite. NBAR can recognize
packets on a complex combination of elds and attributes. However, it is important to
recognize that NBAR is merely a classier, nothing more. NBAR can identify packets that
belong to a certain trafc stream by performing deep-packet inspection, but it is up to the
policy map to determine what should be done with these packets after they have been
identied (in other words, whether they should be marked, policed, dropped, and so on).
NBARs deep-packet classication examines the data payload of stateless protocols and
identies application layer protocols by matching them against a Protocol Description
Language Module (PDLM), which is essentially an application signature. Cisco IOS
software supports 98 protocols via PDLMs as of IOS 12.3. Furthermore, because PDLMs
are modular, they can be added to a system without requiring a Cisco IOS upgrade.
NBAR is dependent on Cisco Express Forwarding (CEF) and performs deep-packet
classication only on the rst packet of a packet stream. The remainder of the packets
belonging to the stream then are CEF-switched. CEF is one of the packet-forwarding
mechanisms within the Cisco IOS Software; there are also fast- and process-switching
forwarding paths.
NOTE
The NBAR classier is triggered by the match protocol command within a class map
denition. It is a more CPU-intensive classier than classiers that match trafc by DSCPs
or ACLs.
NBAR Protocol Classication

NBAR can classify packets based on Layer 4 through Layer 7 protocols, which dynamically
assign TCP/UDP ports. By looking beyond the TCP/UDP port numbers of a packet (known
as subport classication), NBAR examines the packet payload itself and classies packets
on the payload content, such as transaction identiers, message types, or other similar data.
For example, HTTP trafc can be classied by URLs or Multipurpose Internet Mail
Extension (MIME) types using regular expressions within the CLI.
NBAR also can classify Citrix Independent Computing Architecture (ICA) trafc and
can perform subport classication of Citrix trafc based on Citrix published applications.
Requests from Citrix ICA clients can be monitored for a published application that is
destined for a Citrix ICA master browser. After receiving the client requests to the published
Classication Tools
75
application, the Citrix ICA master browser directs the client to the server with the most
available memory. The Citrix ICA client then connects to this Citrix ICA server for the
application.
A summary of protocols that NBAR can use for classication follows. Because new
capabilities are added all the time, this is not an exhaustive list. Not all NBAR classication
involves stateful inspection, and not all match protocol commands trigger NBAR.
Statefully inspected protocols include the following:
FTP
Exchange
HTTP (URL and MIME)
NetShow
RealAudio
r-commands
Oracle SQL*NET
SunRPC
TFTP
StreamWorks
VDOLive
Static protocols include the following:

Exterior Gateway Protocol (EGP)
Generic Routing Encapsulation (GRE)
ICMP
IPinIP
IPSec
EIGRP
BGP
CU-SeeMe
DHCP/BOOTP
Domain Name System (DNS)
Finger
Gopher
HTTP
Secure HTTP (HTTP)
Internet Message Access Protocol (IMPA)
Internet Relay Chat (IRC)
Kerberos
Layer 2 Tunnel Protocol (L2TP)
LDAP
MS-PPTP
MS-SQLServer
NetBIOS
Network File System (NFS)
NNTP
Notes
Network Time Protocol (NTP)
PCAnywhere
POP3
Point-to-Point Tunneling Protocol (PPTP)
RIP
Resource Reservation Protocol (RSVP)
Secure FTP (SFTP)
SHTTP
SIMAP
SIRC
SLDAP
SNNTP
SMTP
SNMP
SOCKS
SPOP3
Secure Shell (SSH)
Secure Telnet (STELNET)
Syslog
Telnet
X Window System
76
Example 3-3 shows the CLI of some NBAR classication congurations.

Example 3-3 NBAR Classication Examples
Router(config)# class-map match-any
Router(config-cmap)# match protocol
ERP
sqlnet
ftp
telnet
Router(config)# class-map match-any AUDIO-VIDEO

Router(config-cmap)# match protocol http mime "*/audio/*"
Router(config-cmap)# match protocol http mime "*/video/*"
Router(config)# class-map match-any WEB-IMAGES
Router(config-cmap)# match protocol http url "*.gif"
Router(config-cmap)# match protocol http url "*.jpg|*.jpeg"
Example 3-3 denes three different class maps. The rst one, the class map ERP, instructs
the classier (NBAR) to pick trafc of any of the protocols listed in the subsequent statements,
which include SQLNET, FTP, or Telnet trafc. In the class map AUDIO-VIDEO, the classier is looking for MIME trafc of particular typesaudio and video, in this case. The last
class map, WEB-IMAGES, is ltering out HTTP trafc for picture (GIF or JPEG) content.
In addition to classication, NBAR can perform protocol discovery using the snifng capabilities of its classication engine. Even if NBAR is not required for QoS policy classication, its protocol-discovery mode can provide valuable information about trafc present on
the network and how much bandwidth each trafc type is using. Such information can be
used in bandwidth provisioning exercises or for capacity planning. An example output of
NBARs protocol-discovery mode is shown in Example 3-4.
Example 3-4 NBAR Protocol Discovery
Router#show ip nbar protocol-discovery stats byte-rate FastEthernet1/0
Protocol
-------------telnet
ftp
http
unknown
Total
Input
30second bit rate
(bps)
-----------------368000
163000
163000
614000
1308000
Output
30second bit rate
(bps)
-----------------0
0
0
0
0
NBAR RTP Payload Classication

Stateful identication of real-time audio and video trafc can differentiate and classify
trafc on the basis of audio and video codec elds within the Real-Time Transport Protocol
(RTP) payload of the packet. Although most voice classication is done in coarser granularity
Marking Tools
77
(by merely separating signaling trafc from speech path [media] trafc) and network access
often is allowed or denied based on the originating port or IP address, sometimes trafc is
desired to be classied by codec. One instance in which this is useful is at the trust boundary
between an enterprise and a service provider network where the SLA is, for example, for
G.729 and G.711 trafc only. In this instance, NBAR can be used to ensure that voice calls
of other codecs are not allowed onto the network.
The same mechanisms can be used if codecs of different bandwidth needs must be ltered out,
for example, to ensure that call admission control (CAC) in the network is not broken. In this
case, low-bandwidth codecs such as G.729 and G.723 can be separated from G.711 trafc.
Filtering trafc by codec can be done by inspecting the payload type (PT) eld within the
RTP header, as dened by the following:
RFC 1889: RTP: A Transport Protocol for Real-Time Applications

RFC 1890: RTP Prole for Audio and Video Conferences with Minimal Control
The command to congure this is as follows:

match protocol rtp [audio | video | payload-type payload-string]
Here, the following is true:
audioSpecies matching by payload-type values 0 to 23

videoSpecies matching by payload-type values 24 to 33
payload-typeSpecies matching by payload-type value, for more granular
matching
For example, the following command instructs NBAR to match RTP trafc with the
payload types 0, 1, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, or 64:
match protocol rtp payload-type "0, 1, 4 - 0x10, 10001b - 10010b, 64"
As shown in the example, the parameters to the match protocol statement can be given in
decimal, hexadecimal (the 0x notation), or binary (the 10001b notation) numbers. Individual
numbers separated by commas can be specied, and ranges of numbers can be used, as in
the case of 4 0x10, which means a decimal value of 4 to a hexadecimal value of 10 (which
equates to a decimal value of 16). Therefore, all RTP payload types between 4 and 16 are
matched for this part of the statement. Similarly, the binary range 10001b to 10010b equates
to 17 to 18 in decimal.
Marking Tools
The main marking tools used today are class-based marking and marking using class-based
policing. Some legacy marking techniques include committed access rate (CAR) and
policy-based routing (PBR). Voice gateway packet marking is another option for IP
telephony applications.
78
Class-Based Marking
Class-based marking, introduced in Cisco IOS Software Release 12.1(2)T, is an MQCbased syntax that uses the set command within a policy map to mark packets, frames, cells,
or labels. Class-based marking was CEF dependent in early Cisco IOS releases (just after
its introduction), but this limitation was listed in subsequent releases soon afterward. If you
are using one of the initial releases, ip cef must be enabled in the global conguration
before using set commands.
Example 3-5 Class-Based Marking Options
Router(config)#policy-map CB-MARKING
Router(config-pmap)#class FOO
Router(config-pmap-c)#set
set ?
atm-clp
Set ATM CLP bit to 1
cos
Set IEEE 802.1Q/ISL class of service/user priority
discard-class Discard behavior identifier
dscp
Set DSCP in IP(v4) and IPv6 packets
fr-de
Set FR DE bit to 1
ip
Set IP specific values
mpls
Set MPLS specific values
precedence
Set precedence in IP(v4) and IPv6 packets
qos-group
Set QoS Group
It is important to remember that class-based marking occurs after classication of the

packet (in other words, set happens after the match criteria). Thus, if used on an output
policy, the packet marking applied can be used by the next-hop node to classify the packet
but cannot be used on this node for classication purposes. On the other hand, if class-based
marking is used on an ingress interface as an input policy, the marking applied to the packet
can be used on the same device on its egress interface for classication purposes.
Another point to note for output policies is that both classication and marking can happen
after tunnel encapsulation, depending on where the service policy is attached. Therefore, if
a policy is attached to a GRE or IPSec tunnel interface, the marking is applied to the original
inner packet header. In most cases, this marking automatically is copied to the tunnel header. On
the other hand, if the policy is attached to the physical interface, only the tunnel header (the
outer header) is marked and the inner packet header is left unchanged.
As an alternative, QoS preclassication, discussed later in this chapter in the section titled
Layer 3 Tunnel Marking Tools, can be used to ensure that classication of the packet
happens on the inner packet header and not the tunnel header values.
Class-Based Policing
Policing and other rate-limiting tools (which are discussed in more detail in Chapter 4,
Policing and Shaping Tools) constitute one of the ways that packets can be marked.
Instead of just marking every packet of a certain type as a particular value, a policer
Marking Tools
79
generally can re-mark (or even drop) packets that violate an SLA. The following command
shows the syntax for a rate-limiter that transmits packets if they conform to a specied rate,
re-marks packets if they exceed the rate, and drops packets if they violate the rate.
police cir 1000000 bc 1000 pir 1000000 be 1000 conform-action transmit
exceed-action set-clp-transmit violate-action drop
Class-based policing can set the IP Precedence, DSCP, MPLS EXP, Frame Relay DE, or
ATM CLP of a packet based on rate-limiting measurements, as shown in Example 3-6.
Example 3-6 Re-Marking Options for the Class-Based Policer
Router(config)#policy-map CB-POLICING
lab-2691(config-pmap-c)#police
police 8000 conform-action ?
drop
drop packet
exceed-action
action when rate is within conform and
conform + exceed burst
set-clp-transmit
set atm clp and send it
set-discard-class-transmit
set discard-class and send it
set-dscp-transmit
set dscp and send it
set-frde-transmit
set FR DE and send it
set-mpls-exp-imposition-transmit set exp at tag imposition and send it
set-mpls-exp-topmost-transmit
set exp on topmost label and send it
set-prec-transmit
rewrite packet precedence and send it
set-qos-transmit
set qos-group and send it
transmit
transmit packet
Committed Access Rate

As with class-based policing, committed access rate (CAR) can be used to set or change
packet markings. However, CAR is an older Cisco IOS policer tool that generally is not
integrated with the MQC syntax and can yield undesirable results if used in conjunction
with service policies. Therefore, CAR is no longer a recommended policer.
Policy-Based Routing
Policy-based routing (PBR) also is an older, non-MQC tool that can perform limited trafc
marking. Although packet marking is not the major function of PBR, it can be used for
writing IP Precedence for packets that match specic criteria.
Voice Gateway Packet Marking

For voice trafc originating on a Cisco voice gateway router, H.323, Media Gateway
Control Protocol (MGCP), and Session Initiation Protocol (SIP) trafc can be marked by
the source gateway. For a long time, only IP Precedence marking was available for VoIP
dial peers, and this only for media (voice) packets. In early releases, ACLs were required
to mark call-signaling packets in conjunction with class-based marking.
80
Cisco IOS Software Release 12.2(2)T introduced the capability to mark voice-sourced
packets on the voice gateway with DSCPs, together with the capability to mark signaling
packets separate from media packets and to mark voice trafc that did not use dial peers
(such as MGCP). The following commands were introduced as part of the simplication of
QoS. They are used for marking the voice trafc at its source, which is more efcient and
easier to manage than manually marking such trafc on the nearest network edge.
H.323 and SIP use a VoIP dial peer command to mark signaling or media packets:
ip qos dscp [af11-af43 | cs1-cs7 | default | ef | num_0-63] [media | signaling]
MGCP uses a global gateway command to mark signaling or media packets:

mgcp ip-tos [rtp | signaling] precedence [0-7]
mgcp ip qos dscp [af11-af43 | cs1-cs7 | default | ef | num_0-63] [media | signaling]
Another move toward simplication in Cisco IOS Software Release 12.2(2)T was to mark
voice and call signaling by default with the appropriate DSCPs. This renders explicit
marking unnecessary unless markings other than the recommended values are desired.
Voice gateway packet-marking features are detailed in Table 3-1.
Table 3-1
Voice Gateway Packet Marking Feature Summary by Cisco IOS Release
Cisco IOS
Release
Protocol
Up to 12.1.5T
SIP, H.323
and 12.2 mainline
12.2.2T and later
SIP, H.323
12.1.5XM and
12.2.2T and later
MGCP
12.2.11T and
later
SIP, H.323
12.2.11T and
later
MGCP
QoS Marking Tools
IP
P
DSCP
Default
Marking
Dial peer for media
Yes
Dial peer, PBR: No
Media: 0
CB marking: Yes
Signaling: 0
PBR, ACL, CB marking for

signaling
Dial peer for media and
signaling marking
Yes
mgcp ip tos for media and

signaling
Yes
Dial peer for media and

signaling marking
Yes
mgcp ip qos dscp for media

and signaling
Yes
Yes
Media: 0
Signaling: 0
No
Media: 5
Signaling: 3
Yes
Media: 5, EF
Signaling: 3,
AF31
Yes
Media: 5, EF
Signaling: 3,
AF31
Marking Tools
81
At the same time, changes were made to the Cisco IP phones and Cisco CallManager to
mark, by default, voice media and signaling packets sourced by these devices. The default
markings are listed in Table 3-2.
Table 3-2
IP Phone and Cisco CallManager Default Voice and Signaling Marking Summary
DSCP
IPP
802.1Q/p CoS
Media
EF
Signaling
AF31 or CS3
Layer 2 Marking Fields

Several cell, frame, or packet elds can be used to carry markings, including the following:
Layer 2 marking elds802.1Q/p CoS bits, MPLS EXP, ATM CLP, and Frame
Relay DE bits
Layer 3 marking eldsIP Precedence or DSCP
Because Cisco Catalyst switches perform scheduling based on Layer 2 802.1Q/p CoS
markings, it is important that Ethernet frames be correctly marked in campus or branch
LANs. However, Layer 2 markings (Ethernet or otherwise) are seldom of end-to-end
signicance. This is because Layer 2 markings are lost whenever the Layer 2 media changes
(for example, from Ethernet to WAN media). In addition, care should be taken that Layer 2
markings are translated to and from Layer 3 markings to ensure consistent end-to-end QoS
for the frame or packet, regardless of where it might travel in the network.
Ethernet 802.1Q/p
Ethernet frames can be marked with their relative importance at Layer 2 by setting the
802.1p User Priority bits (CoS) of the 802.1Q header, as shown in Figure 3-3.
Figure 3-3
Ethernet Frame802.1Q/p CoS Field

Pream.
SFD
DA
SA
Type
TAG
4 Bytes
PT
Three Bits Used for CoS

(802.1p User Priority)
PRI
CFI
VLAN ID
Data
FCS
82
Only 3 bits are available for 802.1p marking. Therefore, only eight classes of service (0
through 7) can be marked on Layer 2 Ethernet frames. These CoS values are identical to IP
Precedence values and typically are assigned according to Table 3-3.
Table 3-3
CoS/IP Precedence Values by Application Types

CoS Value
Application
Reserved
Reserved
Voice
Videoconferencing
Call signaling
High-priority data
Medium-priority data
Best-effort data
The possible values of the 802.1Q/p CoS bits are the same as those for IP Precedence.
Because the eld length is the same, IP Precedence can readily be mapped one to one into
and out of 802.1Q/p CoS values. However, DSCP values (which are 6 bits) cannot be
maintained at the same granularity when mapped into and out of 802.1Q/p CoS values
because some information is lost in the translations.
Ethernet 802.1Q Tunnels

The Cisco Catalyst 3550 switches offer an 802.1Q tunneling feature that enables service
providers to provide Layer 2 VPN tunnels by double-tagging Ethernet frames. As a
tunneling technology, this encapsulates trafc from multiple VLANs of one customer with
a single service provider tag. It preserves the customer VLAN tag over the service provider
network so that the service provider can offer a large number of VLANs to many customers.
Because of the double-tagging of Ethernet frames in 802.1Q tunneling, the CoS value of
the inner frame is not visible to QoS features in the service provider network. Because the
CoS value from the inner frame currently is not copied to the outer frame when the tunnel
is entered, the only form of QoS that the service provider can provide for customer trafc
is QoS on the ingress port, as shown in Example 3-7 (for a Cisco 3550 switch).
Example 3-7 Setting QoS on an Ingress Port of Cisco 3550 Switch
Switchport(config)#interface fastethernet0/1
Switchport(config-if)#mls qos cos 5
! Sets 802.1Q CoS to 5 on outer frame
Switchport(config-if)#mls qos cos override
! Overrides any existing CoS value on the outer frame
Marking Tools
83
Layer 2 protocol packets can be given high priority by using the l2protocol-tunnel cos
global command.
Frame-Relay Discard Eligible Bit

The Frame Relay DE bit in the address eld of a Frame Relay frame is used to indicate
which packets are less important and, therefore, eligible to be dropped before others if
congestion occurs within a Frame Relay cloud. As its name implies, the Frame Relay DE
bit is a single bit that can represent only one of two settings: 0 or 1. If congestion occurs in
a Frame Relay network, frames with the DE bit set at 1 are discarded before frames with
the DE bit set at 0.
Traditionally, Cisco IOS routers could not control the Frame Relay DE bit. The default
Frame Relay DE setting was 0, and only the Frame Relay switch on the service provider
network entry point could set this bit to 1 if the CIR was violated. However, the class-based
marking feature was enhanced in Cisco IOS Software Release 12.2(2)T to allow the router
to control this bit; it provided the option of setting the bit to 1 before trafc exits the router,
and it supported the capability to read the bit upon trafc ingress. Therefore, although the
Frame Relay DE bit is a fairly crude marking option, it can be used in a Frame Relay network
to indicate high-priority trafc (DE bit 0, the default value) and lower-priority trafc (DE
bit 1), which can be dropped should congestion occur. The following is an example of how
the Frame Relay DE bit can be set with class-based marking on trafc that previously was
identied as out-of-contract.
NOTE
In older Cisco IOS releases, class-based marking is dependent on CEF. Therefore, whenever
MQC set commands are to be used, ip cef already must be enabled within the conguration.
In later Cisco IOS releases, this restriction has been lifted.
Example 3-8 shows how the Frame Relay DE bit can be set inside a service policy.
Example 3-8 Setting the Frame Relay DE Bit
Router#show run
policy-map SET-FR-DE
class OUT-OF-SLA
set fr-de
class class-default
fair-queue
84
ATM Cell-Loss Priority Bit

The purpose of the ATM CLP bit is exactly the same as that of the Frame Relay DE bit. It
is a binary eld with two values: 0 (the default), which indicates higher-priority trafc, and 1,
for cells carrying lower-priority trafc that is eligible to be dropped if congestion is
encountered.
Although the capability to set the CLP bit has been available in a policy map since Cisco
IOS Software Release 12.1.5T with the introduction of class-based marking, it is important
to note that not all ATM interface drivers allow this capability. The Cisco 7200 ATM port
adapters (PAs) have long had this capability. The Cisco 2600/3600/3700 ATM interfaces
implemented this capability in Cisco IOS Software Release 12.2.1(1)T, and the digital subscriber line (DSL) interfaces (ADSL and G.SHDSL) require Cisco IOS Software Release
12.2.8YN or later to achieve this feature. Example 3-9 shows how the ATM CLP bit can be
set with class-based marking on trafc that previously was identied as out-of-contract.
Example 3-9 Marking with ATM-CLP
Router# show run
policy-map SET-ATM-CLP
class OUT-OF-SLA
set atm-clp
class class-default
fair-queue
MPLS Experimental Bits

MPLS is a tunneling technology that envelops an IP packet with an MPLS label that has
its own eld denitions for routing and QoS. More than one MPLS label can be used to
envelop a packet. Typically, two labels are used in most MPLS VPN scenarios. In some
scenarios, three labels are used. MPLS labels contain 3 bits for CoS marking. These bits are
referred to as the MPLS EXP bits.
The possible values of the MPLS EXP bits for CoS are the same as those for 802.1Q/p CoS
and IP Precedence. Because of the same length translations (3 bits to/from 6 bits) explained
earlier for 802.1Q/p CoS, IP Precedence (which are 3 bits) readily can be mapped into and
out of MPLS EXP values, but DSCP values (which are 6 bits) cannot be maintained at the
same granularity. Figure 3-4 shows the MPLS EXP bits within an MPLS label.
As of Cisco IOS Software Release 12.1(5)T, the MPLS EXP bits can be read (match command within a class map) and written (set command within a policy map) using MQC.
When a packet enters the MPLS network at the provider edge (PE) router, the IP Precedence
of the packet (by default) automatically is copied to the MPLS EXP eld in the MPLS
header. No explicit action is typically necessary to mark MPLS EXP values, unless the
values require re-marking because of administrative policies.
In theory, upon exiting the MPLS network, the original IP packet re-emerges unchanged
with its IP header type of service (ToS) eld intact. Again, no explicit action needs to be
Marking Tools
85
taken unless the value requires re-marking. While inside the MPLS network, the packets
ToS eld (IP Precedence or DSCP) is irrelevant because the MPLS EXP bits are used to
determine the QoS treatment of the packet within the MPLS cloud, as shown in Figure 3-5.
Figure 3-4
MPLS EXP Bits Within an MPLS Label

Label/Tag
CoS
TTL
MPLS EXP
Label/Tag: 20 Bits
MPLS Experimental (CoS): 3 Bits
Bottom of Stack Indicator (S): 1 Bit
Time-to-Live (TTL): 8 bits
Figure 3-5
Relationship of IP and MPLS Packet Marking
IP Precedence or
DSCP
CE
CE
MPLS EXP
PE
PE
P
V
CE
CE
PE
PE
P
86
In MPLS tunneling scenarios (further discussed in Chapter 16, IPSec VPN QoS Design),
there can be multiple MPLS headers on a packet. To accommodate marking of all or some
of these headers, there are two options on the set mpls experimental command:
set mpls experimental impositionSets a specic value on all labels that are pushed
onto the packet
set mpls experimental topmostSets a specic value only on the topmost MPLS
label on the packet
In practice, however, some service providers currently re-mark the IP Precedence or ToS
elds of packets traversing their MPLS Virtual Private Networks (VPN) to enforce SLAs.
Three main tunneling modes are used for mapping Layer 3 (IP Precedence/DSCP) markings
to and from MPLS EXP values: uniform mode, short-pipe mode, and pipe mode. These
modes are discussed in detail in Chapter 16.
Layer 3 Marking Fields

Layer 3 packet marking with IP Precedence and DSCPs is the most widely deployed
marking option because Layer 3 packet markings have end-to-end network signicance and
easily can be translated to the Layer 2 frame markings previously discussed.
As with Layer 2 tunneling, Layer 3 tunneling technologies pose a challenge in preserving
packet markings by enveloping the packet with a new header/packet. Some technologies
automatically copy the inner packet ToS eld to the outer header packet, whereas others
do not.
IP Type of Service and IP Precedence

The second byte in an IPv4 packet is the type of service (ToS) byte. The rst 3 bits (by
themselves) are referred to as the IP Precedence bits, as shown in Figure 3-6.
Figure 3-6
IPv4 Type of Service Byte (IP Precedence Bits and DSCP)

ToS
Byte
Version
Length
Len
ID
Offset
TTL
Proto
FCS
IP SA
IP DA
Data
IPv4 Packet
7
IP Precedence
Standard
Standard IPP
IPv4
Unused
DiffServ Code Point (DSCP)
IP ECN
DiffServ
DiffServ Extensions
Extensions
The IP Precedence bits, similar to the 802.1Q/p CoS bits and the MPLS EXP bits, allow for
only eight values of marking (0 through 7). Because values 6 and 7 generally are reserved
Marking Tools
87
for network control trafc (such as routing) and value 0 is the default marking value, really
only ve remaining values can be used to differentiate non-best-effort trafc. Of these ve
remaining values, however, the following is true:
IP Precedence value 5 is recommended for voice.

IP Precedence value 4 is shared by videoconferencing and streaming video.
IP Precedence value 3 is recommended for call signaling.
This leaves only two marking values (IP Precedence 1 and 2) available for all data
application marking options. Thus, many enterprises nd IP Precedence marking to be
overly restrictive and favor instead the 6-bit/64-value DSCP marking model.
NOTE
In this book, IP Precedence is viewed as a legacy technology, and all Layer 3 marking
recommendations are based on DSCP only (unless specic constraints exist).
Differentiated Services Code Points

As shown in Figure 3-6, DSCPs use the same 3 bits as IP Precedence and combine these
with the next 3 bits of the ToS byte to provide a 6-bit eld for QoS marking. Thus, DSCP
values range from 0 (000000) to 63 (111111). This range provides unprecedented richness
in marking granularity.
DSCP values can be expressed in numeric form or by special keyword names, called
per-hop behaviors (PHB). Three dened classes of DSCP PHBs exist: Best-Effort (BE
or DSCP 0), Assured Forwarding (AFxy), and Expedited Forwarding (EF). In addition to
these three dened PHBs, Class-Selector (CSx) codepoints have been dened to be backward
compatible with IP Precedence (in other words, CS1 through CS7 are identical to IP
Precedence values 1 through 7). The RFCs describing these PHBs are 2547, 2597, and
3246.
RFC 2597 denes four Assured Forwarding classes, denoted by the letters AF followed by
two digits. The rst digit denotes the AF class and can range from 1 through 4. (Incidentally,
these values correspond to the three most signicant bits of the codepoint, or the IPP value
that the codepoint falls under.) The second digit refers to the level of drop preference within
each AF class and can range from 1 (lowest drop preference) to 3 (highest drop preference).
For example, during periods of congestion (on an RFC 2597compliant node), AF33 would
be dropped more often (statistically) than AF32, which, in turn, would be dropped more
often (statistically) than AF31. Figure 3-7 shows the Assured Forwarding PHB encoding
scheme.
88
Figure 3-7
DiffServ Assured Forwarding PHB Encoding Scheme

IP Header ToS Byte
DSCP
AFxy
X
Class
Drop Precedence
Figure 3-8 shows a summary of PHBs along with their decimal and binary equivalents.
Figure 3-8
DiffServ PHBs with Decimal and Binary Equivalents
DiffServ Code Points

(DSCP)
Per-Hop Behaviors (PHB)
46
101110
EF
Expedited Forwarding
Assured Forwarding
Low Drop
Pref
Med Drop
Pref
High Drop
Pref
Class 1
AF11
AF12
AF13
001010 001100 001110
Class 2
AF21
AF22
AF23
010010 010100 010110
Class 3
AF31
AF32
AF33
011010 011100 011110
Class 4
AF41
AF42
AF43
100010 100100 100110
10
18
26
34
12
20
28
36
14
22
30
38
Best Effort
BE
000000
Layer 3 Tunnel Marking Tools

Cisco routers offer a variety of tunneling features, such as GRE, IPSec, and L2TP, which
enable service providers to provide Layer 3 VPN tunnels by enveloping one IP packet
within another. Such encapsulation masks the original header information to provide
features such as privacy, encryption, and address preservation. Tunneling technologies also
are used to carry non-IP protocols over an IP backbone.
Marking Tools
89
A wide range of packet header layouts with tunneling technologies exist, but the primary
characteristic that they have in common is that the original IP header is enveloped in an
outer header packet. While in the tunnel, only the outer IP headers ToS byte is examined
to determine what QoS policies should be applied to the packet. The ToS byte from the
inner packet might or might not be copied automatically to the outer header packet. If it is
not copied automatically, explicit commands are required to copy the ToS byte (or to set the
outer header ToS byte, independent of the inner packets ToS values).
Some example packet header layouts of GRE and IPSec packets are shown in Figure 3-9.
Figure 3-9
L3 Tunnel Packet Layout Examples

GRE Tunneling
GRE
IP
Hdr
UDP
RTP Voice
IPSec ESP Transport Mode

IPSec ESP ESP
GRE
Hdr Hdr
IV
IP
Hdr
UDP
RTP Voice
ESP
ESP
Pad/NH Auth
IPSec ESP Tunnel Mode

IPSec ESP ESP GRE IP
GRE
Hdr
Hdr Hdr
IV
IP
Hdr
UDP
RTP Voice
ESP
ESP
Pad/NH Auth
Three methods provide QoS marking for Layer 3 tunnels: QoS preclassication (QoS for
VPNs feature), ToS copying/reection, and independent header-packet marking. Each is
discussed in more detail in the following sections.
QoS Preclassify
The QoS preclassify feature was introduced in Cisco IOS Software Release 12.1(5)T on
Cisco 7100 and 7200 series routers and in Cisco IOS Software Release 12.2(2)T for lowerend routers. This command creates a clone of the inner packet header (strictly for internal
router processing) before the packet is enveloped. Upon egress, the router compares the
cloned header against any policies applied to the egress interface (because it no longer can
read information from the original packet header because it is enveloped). Then the applicable policies are serviced on the packet ow, and the clone is discarded. An advantage of
the QoS preclassify feature is that not only is the ToS byte of the inner header used for QoS
classication purposes, but other IP/TCP/UDP header parameters such as source/destination IP addresses and source/destination ports can be used.
90
Strictly speaking, the QoS preclassify feature is only a classication feature. Its marking
functionality is only transient, in the sense that it makes a copy of the inner packet header
and its markings, but this header never is transmitted as part of the packet.
Examples of the qos pre-classication command for various types of tunnels are shown in
Example 3-10.
Example 3-10 QoS Preclassication Examples
GRE and IPIP Tunnels
Router(config)# interface tunnel0
Router(config-if)# qos pre-classify
L2F and L2TP Tunnels:
Router(config)# interface virtual-template1
Router(config-if)# qos pre-classify
IPsec Tunnels:
Router(config)# crypto map secured-partner-X
Router(config-crypto-map)# qos pre-classify
ToS Reection
QoS marking for tunnels also can be achieved by copying the ToS byte from the inner
header to the outer header. This is done by default on most platforms for IPSec and GRE
tunnels. For L2TP, the l2tp tos reect command can be used.
Independent Header-Packet Marking

Another option is to mark the tunnel header explicitly as any other packet would be marked.
This might be the least useful of the tunnel-marking methods because the characteristics for
QoS treatment almost always are associated with the inner packet. Nevertheless, it is
possible to mark the tunnel header independently with IP Precedence or DSCPs.
Translating Layer 2 and Layer 3 Packet Markings

The Layer 2 and Layer 3 marking elds discussed in the previous sections are summarized
in Table 3-4.
It is important to remember that several technologies change packet headers or wrap one
packet into another outer packet so that one packet or frame becomes the payload of the
next. When this happens, packet marking is lost unless it explicitly is carried forward to the
new packet (or frame) header. These repacketization changes occur when a data segment
crosses a Layer 3 or Layer 2 technology boundary or when tunneling technologies are used.
To preserve packet markings end to end, there is often the need to translate one type of
marking to another at a network boundary (for example, LAN to WAN edge) or technology
boundary (for example, the start of an encryption tunnel between two sites).
Marking Tools
Table 3-4
91
L2 and L3 Marking Options Summary

Technology
Layer
Marking Field
Field Width (Bits)
Value Range
Ethernet
802.1Q/p
0 to 7
Frame Relay
DE bit
0 to 1
ATM
CLP bit
0 to 1
MPLS
EXP
0 to 7
IP
IP Precedence
0 to 7
IP
DSCP
0 to 63
Some examples include these:
VoIP over Frame RelayA translation from Layer 3 to Layer 2 in which a VoIP
packet is enveloped within a Frame Relay frame. The Frame Relay frame header
marking eld (DE bit) is 0 unless it is marked explicitly.
Recommendation: Leave the voice packets DE bit as 0, but consider marking lowpriority data packets sharing the same congestion points with DE bit 1.
VoIP over ATMA translation from Layer 3 to Layer 2 translation in which a VoIP
packet is enveloped within multiple ATM cells (typically AAL5). The ATM cell
header marking eld (CLP) should be clear.
Recommendation: Leave the voice packets ATM CLP as 0, but consider marking lowpriority data packets that share the same congestion points with CLP 1.
VoIP over Ethernet to VoIP over a WANA translation from Layer 2 to Layer 3 in
which VoIP on a LAN segment carries an 802.1Q/p packet header marking. When the
packet hits a router and heads out over the WAN, the Layer 3 IP packet containing
the voice payload might or might not be marked appropriately, depending on the
conguration and capabilities of the switch, router, or IP phone.
Recommendation: Congure that LAN switch to convert 802.1Q/p marking to DSCPs
if the packet is handed off to a Layer 3 segment. If the switch is not capable of such
mapping, perform the mapping from Layer 2 to Layer 3 on the routers LAN edge. A
mapping from Layer 3 to Layer 2 also might be needed on remote-branch routers to
restore lost CoS mappings for VoIP Ethernet frames entering the branch from the WAN.
VoIP over MPLSA translation of Layer 3 to Layer 2. As with other tunneling

technologies, MPLS envelops the IP packet with another header (MPLS label). On
tunnel entry, the IP packets ToS field is mapped to the MPLS EXP bits by default.
Recommendation: Ensure that the default mapping feature has been implemented in
the Cisco IOS software release and platform; otherwise, mark the MPLS EXP eld
explicitly. Keep in mind that the MPLS EXP eld is only 3 bits long, so IP Precedence
will translate correctly, but DSCPs will lose granularity in the translation(s). Many
92
enterprise networks do not have control over the MPLS backbone they might use. If
so, work with the service provider offering the MPLS network to ensure that the
network is congured correctly.
Tunnel technologies such as L2TP, IPSec, and GREA translation of Layer 3 to

Layer 3. These technologies wrap an IP packet inside another IP packet by putting a
tunnel header in the front of the packet. Aside from the fact that there are bandwidth
provisioning implications with such additional overhead, this masks the packet header
marking of the inner packet. Note that this situation is potentially problematic only
upon entering the tunnel because a new packet header is added to the existing packet.
Upon exiting the tunnel, the original packet re-emerges with its marking intact, so no
extra action or caution is necessary.
Recommendation: Use the QoS preclassify feature to ensure that packet classication
happens on the inner packet.
802.1Q/p to and from DSCP

Figure 3-10 shows an example of how Layer 2 (802.1Q/p CoS) markings can be translated
to Layer 3 (DSCP) markings using class-based marking. In this example, CoS 5 is mapped
to and from DSCP EF, and CoS 3 is mapped to and from DSCP CS3. (These are the typical
values used for voice and call signaling for IP telephony.) Cisco IP phones mark voice
packets to CoS 5 and DSCP EF, and call signaling packets to CoS 3 and DSCP CS3 or AF31
automatically and by default (rendering such mapping of Layer 2 to Layer 3 unnecessary,
in most cases).
NOTE
Ethernet 802.1Q/p is the only Layer 2 marking technology that might require bidirectional
mappings (Layer 2 to Layer 3 and Layer 3 to Layer 2). Cisco Catalyst switches (including
those at remote branch locations) assign scheduling based on Layer 2 802.1p CoS markings,
which are lost when the packets traverse a WAN media. All other Layer 2 marking options
are applicable to the WAN/VPN transit cloud only and lose their relevance after the frame
is received at the remote branch. Because of this, and because the underlying Layer 3
markings are preserved through the transit cloud, a second mapping is rarely necessary with
Frame Relay DE, ATM CLP, and MPLS EXP markings.
Marking Tools
93
Figure 3-10 LAN-to-WAN Mapping of CoS and DSCP
Input LAN Interface

class-map COS5
match cos 5
WAN
class-map COS3
match cos 3
!
policy-map COS-TO-DSCP
class COS5
set ip dscp ef
fa0/1.100
class COS3
set ip dscp af31
class class-default
set ip dscp 0
!
interface FastEthernet0/1.100
service-policy input COS-TO-DSCP
Output LAN Interface

class-map DSCP-EF
match ip dscp ef
class-map DSCP-AF31
match ip dscp af31
!
policy-map DSCP-TO-COS
class DSCP-EF
set cos 5
class DSCP-AF31
set cos 3
class class-default
set cos 0
!
service-policy output DSCP-TO-COS
Example 3-11 shows how the policy maps in Figure 3-10 can be applied to outgoing Voice
VLAN and Data VLAN FastEthernet 802.1Q subinterfaces on the router.
Example 3-11 Applying L3-to-L2 Marking on LAN Interface
Router#sh run
interface FastEthernet0/1
no ip address
full-duplex
!
description Voice-VLAN
encapsulation dot1Q 100
ip address 10.6.0.129 255.255.255.192
!
description DATA-VLAN
ip address 10.6.0.1 255.255.255.128
!
94
DSCP to Frame Relay DE Bit

Figure 3-11 shows an example of using the Frame Relay DE bit to preserve some level of
priority in the Frame Relay cloud. Within this enterprise, scavenger trafc is marked to
DSCP CS1. If congestion occurs within the Frame Relay cloud, such trafc should be the
rst to be dropped. On the routers egress interface, all frames carrying scavenger trafc are
to have their Frame Relay DE bits set to 1. Furthermore, all other trafc is rate limited, and
frames of trafc that exceed this limit also have their Frame Relay DE set to 1.
Figure 3-11 Trafc Priority Marking with Frame Relay DE Bits
class-map SCAVENGER
match ip dscp cs1
!
policy-map FR-OUT
class SCAVENGER
set fr-de
class class-default
police 512000 1500 1500
conform-action transmit
exceed-action set-frde-transmit
Frame Relay
Cloud
DSCP to ATM CLP Bit

Figure 3-12 shows an example of using the ATM-CLP bit to preserve some level of priority
in the ATM cloud. As in the previous example, scavenger trafc is marked to DSCP CS1.
If congestion occurs within the ATM cloud, such trafc should be the rst to be dropped.
On the routers egress interface, all frames carrying scavenger trafc are to have their ATM
CLP bits set to 1. Furthermore, all other trafc is being rate limited, and cells of trafc that
exceed this limit also have their ATM CLP bits set to 1.
Figure 3-12 Trafc Priority Marking with ATM CLP Bits
class-map SCAVENGER
match ip dscp cs1
!
policy-map ATM-OUT
class SCAVENGER
set atm-clp
class class-default
police 512000 1500 1500
exceed-action set-clp-transmit
ATM
Cloud
Marking Tools
95
DSCP to MPLS EXP Bits

Figure 3-13 shows an example of mapping DSCPs to MPLS EXPs. This might be needed
when MPLS VPN service providers offer various levels of service based on MPLS EXP
markings. Currently, though, most service providers base their admission to various levels
of service by examining the DSCP markings of packets offered to them from their enterprise
customer edge (CE) routers. In this example, the service provider is offering three levels of
service: Realtime (as admitted by MPLS EXP value 5), Business-Data (as admitted by
MPLS EXP value 3), and Best Effort (everything else). The CE-to-PE link in this example
is a T1 and, as such, has no serialization issues (which are discussed in greater detail later).
The enterprise customer wants both voice and call-signaling trafc to be admitted to the
service providers Realtime class. Therefore, the customer maps both DSCP EF and DSCP
AF31 to MPLS EXP 5.
By default, voice automatically would have been mapped from DSCP EF to MPLS EXP 5.
However, call signaling would have been mapped to MPLS EXP 3 by default. Furthermore,
the enterprise customer has transactional data marked to DSCP AF21 and bulk data marked
to DSCP AF11, which, by default, would be mapped to MPLS EXP 2 and 1, respectively.
The enterprise customer wants both of these to be admitted to the service providers
Business-Data class. To accomplish this, the enterprise customer manually maps DSCP
AF21 and AF11 to MPLS EXP 3. Everything else is marked to MPLS EXP 0.
Figure 3-13 illustrates how and where the mapping of the DSCP to MPLS EXP value could
occur, in the case of a service-provider managed CE scenario: specically, under a Pipe
Mode with Explicit Null LSP conguration (for more detail on this design option, refer to
Chapter 15 MPLS VPN QoS Design).
However, in most scenarios, enterprise customers have no control or visibility into the
MPLS backbone, which typically is owned and managed by the service provider.
IP Precedence to ATM/Frame Relay PVCs (PVC Bundling)

Under some circumstances, multiple permanent virtual circuit (PVC) models might be
economically attractive to enterprise customers. Such multiple-PVC models offer enterprise
customers more granular levels of service across ATM or Frame Relay clouds than simple
CLP or DE bit markings alone. When multiple PVCs exist, enterprises can use PVC
bundles to assign relative trafc priorities over these WAN topologies.
NOTE
Although bundling is widely deployed, it is an aging and inefcient QoS technology. At the
time of this writing, it supports only IP Precedence, not DSCP. Bundling is inefcient
because lower-priority applications never gain access to any excess bandwidth that might
exist on higher-priority PVCs. Therefore, any unused bandwidth on these PVCs is wasted.
96
Figure 3-13 Trafc Priority Marking with MPLS EXP Bits

class-map VOICE
match ip dscp ef
class-map CALL-SIGNALING
match ip dscp af31
class-map TRANSACTIONAL-DATA
match ip dscp af21
class-map BULK-DATA
match ip dscp af11
!
policy-map MPLS-OUT
class VOICE
set mpls experimental 5
class CALL-SIGNALING
class TRANSACTIONAL-DATA
class BULK-DATA
class class-default
MPLS
VPN
CE
PE
An example of bundling is outlined in Figure 3-14. An enterprise has purchased four separate
ATM PVCs with varying levels of ATM QoS. It wants voice (IP Precedence 5) to be assigned
to a dedicated variable bit rate real-time (VBR-rt) PVC, video (IP Precedence 4) and call
signaling (IP Precedence 3) to be assigned to a variable bit rate non-real-time (VBR-nrt)
PVC, transactional data (IP Precedence 2) and bulk data (IP Precedence 1) to be assigned
to an available bit rate (ABR) PVC, and everything else to be assigned to an unspecied bit
rate (UBR) PVC.
Figure 3-14 IP Precedence to ATM PVC Bundle Example
ATM PVC
Bundle
IPP 5
VBR-rt PVC
IPP 3-4
VBR-nrt PVC
IPP 1-2
ABR PVC
IPP 0
UBR PVC
Single ATM Interface With Multiple VCs
A sample ATM PVC bundling conguration that corresponds to the example is shown in
Example 3-12.
Marking Tools
97
IP Precedence-to-ATM VC bundling has been a Cisco IOS feature for several years.
Bundling functionality for Frame Relay PVCs was introduced in Cisco IOS Software
Release 12.2.1(3)T. Mapping IP Precedence markings to ATM VCs provides truer levels of
service because of the ATM service class attributes that dene the ATM PVCs. Frame Relay
PVCs have no intrinsic service class attributes associated with them, but they do offer the
capability to guarantee bandwidth to a particular class of trafc across the backbone.
Example 3-12 Sample ATM PVC Bundling Conguration
Router# show run
vc-class atm VOICE-PVC-256
vbr-rt 256 256
tx-ring-limit 3
precedence 5
no bump traffic
protect group
!
vc-class atm VIDEO-PVC-256
vbr-nrt 256 256
tx-ring-limit 3
precedence 4-3
no bump traffic
protect group
!
vc-class atm BUSINESS-DATA-PVC-512
abr 512 512
precedence 2-1
no bump traffic
protect group
!
vc-class atm BEST-EFFORT-PVC-512
ubr 512
tx-ring-limit 3
precedence other
VC bundling offers QoS by separating classes of trafc over individual PVCs. Therefore,
it is important to remember that other QoS tools targeted at prioritizing different types of
trafc on the same VC, such as LLQ, do not readily apply here. Also, PVC bundles do not
offer bandwidth-sharing arrangements (such as Multilink Point-to-Point Protocol [MLP]
and Frame Relay multilink bundling) because they dedicate a particular PVC to a given
class of trafc. If that class does not use its bandwidth allocation, it cannot be reallocated
to other types of trafc. If bandwidth-sharing features are required, Multilink PPP over
ATM (MLPoATM) or Multilink PPP over Frame Relay (MLPoFR) bundles must be used
in conjunction with MQC-based LLQ/CBWFQ policies.
98
Table Map Feature

Although the set command can be used individually, as discussed in the previous sections,
to translate a packet marking from one type to another, this might be cumbersome in the
conguration if the same translation is required in many places. To ease the conguration
of translating packet markings, the table map feature can be used. The command syntax is
as follows:
table-map table-map-name map from from-value to to-value
[default default-action-or-value]
This can be used on the set command as shown in Example 3-13.

Example 3-13 Conguring the Table Map Feature
Router(config)#table-map table1
Router(config-tablemap)#map from 2 to 1
Router(config)#policy-map CB-marking
Router(config-pmap-c)#set mpls experimental topmost qos-group table table1
Example 3-14 shows a number of set command examples using the table map feature to
translate from one type of packet marking to another.
Example 3-14 Use of the Table Map Feature
set
set
set
set
set
set
set
set
set
set
set
set
precedence cos table table-map-name

dscp cos table table-map-name
cos precedence table table-map-name
cos dscp table table-map-name
qos-group precedence table table-map-name
qos-group dscp table table-map-name
mpls experimental topmost qos-group table table-map-name
mpls experimental imposition precedence table table-map-name
mpls experimental imposition dscp table table-map-name
qos-group mpls exp topmost table table-map-name
precedence qos-group table table-map-name
dscp qos-group table table-map-name
Summary
This chapter examined classication and marking features and tools. Classication is the
action of inspecting a packet (certain elds within the packet) to determine what type of
packet or trafc it is. This determination is used to guide the treatment that the packet (and
other packets of the same trafc type or stream) will receive from the node and the network.
Marking is the action of changing a eld within the packet header to note the determination
reached by the classier. The various ways of doing packet marking at L2 and L3 up to L7
were illustrated, and ways to translate one type of marking to another were discussed.
The treatment of the packet, which is based on the classication and marking results, includes
capabilities such as policing, shaping, and queuing. Policing and shaping are discussed
Further Reading
99
in Chapter 4, Policing and Shaping Tools, and queuing is discussed in Chapter 5,

Congestion-Management Tools.
Further Reading
General
Class-based marking: http://www.cisco.com/univercd/cc/td/doc/product/software/

ios122/122cgcr/fqos_c/fqcprt1/qcfcbmrk.htm.
Class-based policing (Cisco IOS Software Release 12.2.2T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122cgcr/fqos_c/fqcprt4/qcfpoli.htm.
Frame Relay DE bit marking (Cisco IOS Software Release 12.2.2T): http://
www.cisco.com/univercd/cc/td/doc/product/software/ios121/121newft/121t/121t5/
cbpmark2.htm#1037921.
Enhanced packet marking (Cisco IOS Software Release 12.2.1[3]T):

http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/122newft/
122t/122t13/ftenpkmk.htm.
Packet classication based on Layer 3 packet length (Cisco IOS Software Release
12.2.13T): http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/
122newft/122t/122t13/ftmchpkt.htm.
Packet classication using the Frame Relay DLCI number (Cisco IOS Software
Release 12.2.13T): http://www.cisco.com/univercd/cc/td/doc/product/software/
ios122/122newft/122t/122t13/ftpcdlci.htm.
DiffServ for end-to-end quality of service (Cisco IOS Software Release 12.1.5T):
http://www.cisco.com/univercd/cc/td/doc/product/software/ios121/121newft/121t/
121t5/dtdfsv.htm.
Classifying VoIP signaling and media with DSCP for QoS (Cisco IOS Software
ios122/122newft/122t/122t2/ft_dscp.htm.
Control plane DSCP support for RSVP (Cisco IOS Software Release 12.2.2T):
122t2/dscprsvp.htm.
Voice Gateway Packet Marking (Cisco IOS Software Release 12.2.2T):

122t/122t2/ft_dscp.htm.
DiffServ
100
L2 Protocol Tunneling
Catalyst 3550 IOS (Cisco IOS Software Release 12.1.14EA1) documentation:

http://www.cisco.com/univercd/cc/td/doc/product/lan/c3550/12114ea1/3550scg/
swtunnel.htm.
Catalyst 3550 802.1Q Tunneling Conguration Guide: http://wwwin.cisco.com/eag/

dsbu/solutions/documents/06_802.1Q%20Tunneling%20Cong%20Guide.doc.
L2TP IP ToS reect command IOS (Cisco IOS Software Release 12.3) documentation:
http://www.cisco.com/univercd/cc/td/doc/product/software/ios123/123cgcr/dial_r/
dia_l1g.htm#1131064.
Quality of service for Virtual Private Networks (Cisco IOS Software Release 12.1.5T):
121t5/dtqosvpn.htm.
Quality of service for Virtual Private Networks (Cisco IOS Software Release 12.2.2T):
122t2/ftqosvpn.htm.
Network-Based Application Recognition and Distributed Network-Based Application

Recognition (Cisco IOS Software Release 12.1.5T): http://www.cisco.com/univercd/
cc/td/doc/product/software/ios122/122newft/122t/122t8/dtnbarad.htm.
NBAR RTP Payload Classication (Cisco IOS Software Release 12.2.8T):

122t8/dtnbarad.htm.
Network-Based Application Recognition Protocol Discovery Management

Information Base (Cisco IOS Software Release 12.2.1[5]T): http://www.cisco.com/
univercd/cc/td/doc/product/software/ios122/122newft/122t/122t15/ftpdmib.htm.
MPLS class of service enhancements (Cisco IOS Software Release 12.1.5T):

121t5/mct1214t.htm.
MPLS QoS multi-VC mode for PA-A3 (Cisco IOS Software Release 12.2.2T):
122t2/cos1221t.htm.
VPN
NBAR
MPLS
Further Reading
101
DiffServ-aware MPLS trafc engineering (DS-TE) (Cisco IOS Software Release

122newft/122t/122t4/ft_ds_te.htm.
MPLS DiffServ-aware trafc engineering (DS-TE) over ATM (Cisco IOS Software
ios122/122newft/122t/122t8/ft_ds_te.htm.
IPATM/Frame Relay Bundles
IP to ATM class of service (Cisco IOS Software Release 12.0.3T):

120t/120t3/ipatmcs2.htm.
IP to ATM CoS, per VC WFQ and CBWFQ (Cisco IOS Software Release 12.0.5T):
120t5/ipatm3.htm.
IP to ATM class of service mapping for SVC bundles (Cisco IOS Software Release
122newft/122t/122t4/ftsvbund.htm.
MPLS EXP to ATM VC bundling (Cisco IOS Software Release 12.2.8T):

122t8/ftmpls.htm.
Frame Relay PVC bundles with QoS support for IP and MPLS (Cisco IOS Software
ios122/122newft/122t/122t13/ft_frbnd.htm.
MPLS EXP to Frame Relay VC bundling (Cisco IOS Software Release 12.2.13T):
122t13/ft_frbnd.htm.
Level 2 to Level 3 Packet-Marking Translation
Enhanced packet marking (Cisco IOS Software Release 12.2.1[3]T):

122t13/ftenpkmk.htm.
Policing, shaping, and the differences between the intent and operation of these
techniques
Policing tools such as committed access rate (CAR) and class-based policing
Shaping tools such as ATM and Frame Relay trafc shaping (FRTS), generic trafc
shaping (GTS), and class-based shaping
Advanced policing topics, such as the single-rate and two-rate three-color policers,
and hierarchical, multiaction, color-aware, and percentage-based policing
CHAPTER

Policers and shapers are the oldest forms of QoS mechanisms. These tools have similar
objectivesnamely, to identify and respond to trafc violations. Policers and shapers
usually identify trafc violations in an identical manner. However, they differ in how they
respond to the violations.
Policers perform instantaneous checks for trafc violations and take immediate prescribed
actions when such violations occur. For example, a policer can determine whether the
offered load is in excess of the dened trafc rate and then can re-mark or drop the outof-contract trafc. Figure 4-1 shows only a dropping action taken by the policer, although
other types of actions (for example, re-marking the packet or simply transmitting it) are also
possible.
Shapers are trafc-smoothing tools that work in conjunction with queuing mechanisms.
The objective of a shaper is to send all the trafc offered to an interface, but to smooth it
out so that it never exceeds a given rate. Shapers usually are employed to meet service-level
agreements (SLA) or to compensate for nonbroadcast, multiple-access idiosyncrasies. If
the offered trafc momentarily exceeds the dened rate, the excess trafc is buffered and
delayed until the offered trafc once again dips below the dened rate.
Figure 4-1 illustrates the difference between policing and shaping. Both mechanisms
measure trafc against a given trafc rate.
Chapter 4: Policing and Shaping Tools
Generic Policing Versus Shaping
Offered Traffic
Figure 4-1
Policing
Offered Traffic
104
Time
Shaping
Offered Traffic
Time
Offered Traffic
Traffic Rate (Sent)
Time
Traffic Rate (Sent)
Time
Table 4-1 compares the characteristics of policing and shaping tools.

Table 4-1
Comparisons Between Policers and Shapers

Policer
Shaper
Causes TCP resends as trafc is dropped
Typically delays (rather than drops) trafc;

involves fewer TCP resends
Inexible and inadaptable; makes instantaneous

packet drop decisions
Can adapt to network congestion by queuing

excess trafc
An ingress or egress interface tool
Typically an egress interface tool
Rate limiting without buffering
Rate limiting with buffering
Although policing and shaping tools are not employed directly to provide QoS for real-time
packets (such as voice and interactive video), they do regulate and stabilize trafc ows so
that service guarantees can be made for such real-time applications. Without policers and
shapers, unexpected bursts in data trafc could affect the jitter and latency thresholds
adversely for real-time trafc.
Also of note is that policers and shapers can work in tandem; they are not mutually
exclusive tools.
Token Bucket Algorithms
105
Token Bucket Algorithms

Cisco IOS policers and shapers are modeled after token bucket algorithms. Essentially,
token bucket algorithms are metering engines that keep track of how much trafc can be
sent to conform to the specied trafc rates. A token permits the algorithm to send a single
bit (or, in some cases, a byte) of trafc. These tokens are granted at the beginning of some
time increment, typically every second, according to the specied rate referred to as the
committed information rate (CIR). The CIR is the access bit rate contracted with a service
provider or the service level to be maintained.
For example, if the CIR is set to 8000 bps, then 8000 tokens are placed in a bucket at the
beginning of the time period. (Note that this description represents a simplied view of the
algorithm and might not be strictly true in all cases, but it illustrates the general operation
of the policing mechanism.) Each time a bit of trafc is offered to the policer, the bucket is
checked for tokens. If there are tokens in the bucket, the trafc is passed. One token is
removed from the bucket for each bit of trafc that is passed. Therefore, trafc is viewed to
conform the rate, and the specied action for conforming trafc is taken. (Typically, the
conforming trafc is transmitted.) When the bucket runs out of tokens, any additional
offered trafc is viewed to exceed the rate, and the exceed action is taken. (The exceeding
trafc typically either is re-marked or is dropped.)
NOTE
At the end of the second, there might be unused tokens. The handling of the unused tokens
is a key differentiator among policers. This is discussed in the Policers section later in
this chapter.
Because the clock rate of the interface cannot change to enforce a policy, the only way that
a rate limit can be imposed on an interface is to use time-division multiplexing (TDM). With
TDM, when a rate limit (or CIR) is imposed on an interface, the limited trafc is allocated
a subsecond time slice during which it can be sent. This subsecond time slice is referred to
as the interval (or Tc). For example, if an 8-kbps CIR is imposed on a 64-kbps link, trafc
can be sent for an interval of 125 ms (64,000 bps / 8000 bits).
The entire amount of the CIR (8000 bits) could be sent at once, but then the algorithm
would have to wait 875 ms before it could send any more data (to impose the rate limit).
Such an interpacket delay likely would be viewed as excessive. Therefore, to smooth out
the ow over each second, the CIR is divided into smaller units, referred to as the committed
burst (Bc), which is the sustained number of bits that can be transmitted per interval. These
smaller units are sent over multiple instances during a single second. Continuing with the
previous example, if the Bc is set to 1000, each committed burst can take only 15.6 ms
(1000 bits / 64,000 bps) to send trafc out the interface at the clock rate. The algorithm
waits 109.4 ms (125 ms 15.6 ms) and sends another 15.6 ms of data. This process is
repeated a total of eight times during the second.
106
Therefore, the token bucket algorithm is as follows:

CIR = Bc / Tc
Cisco IOS Software does not permit the explicit denition of the interval. Instead, it takes
the CIR and Bc as arguments from which the interval and the number of bursts per second
are derived. For example, if the CIR is 8000 and the Bc is set to 4000, two bursts occur per
second (Tc = 500 ms). If the Bc is set to 2000, four bursts occur per second (Tc = 250 ms).
If the Bc is set to 1000, eight bursts occur per second (Tc = 125 ms).
The preceding example illustrates the operation of the feature from a theoretical perspective. From a practical perspective, when implementing networks, Tc should not exceed
125 ms. Shorter intervals can be congured and are necessary to limit jitter in real-time trafc,
but longer intervals are not practical for most networks because the interpacket delay
becomes too large.
The earliest policers all use a single-rate two-color marker/policer model with a single
token bucket algorithm. In this model, trafc is identied as one of two states (colors):
conforming to or exceeding the CIR. Marking and dropping actions are performed on each
of these two states of trafc. This type of marker/policer is fairly crude and is illustrated in
Figure 4-2.
Single
Token Bucket
Policer
Time
Offered Traffic
Single-Rate Two-Color Policer Effect on Trafc Flow (Single Token Bucket Policer)
Offered Traffic
Figure 4-2
Traffic Rate (Sent)
Time
Although a policer can be deployed at ingress or egress interfaces, it often is deployed at

the network edge on trafc ingress. If packets will be dropped, there is little point in
spending valuable CPU cycles routing and processing these packets. However, policers also
often are deployed at the trafc egress interface to control bandwidth used or are allocated
to a particular class of trafc. This bandwidth decision often is not made until the packets
reach the egress interface.
Policers
107
Policers
As mentioned before, policers determine whether each packet conforms or exceeds (or,
optionally, violates) to the trafc congured policies and take the prescribed action. The
action taken can include dropping or re-marking the packet. Conforming trafc is trafc
that falls within the rate congured for the policer. Exceeding trafc is trafc that is above
the policer rate but still within the burst parameters specied. Violating trafc is trafc that
is above both the congured trafc rate and the burst parameters.
How trafc is separated, where it is policed, and why are important questions to keep in
mind in the overall network QoS design. It is not productive, for example, to police DSCP
EF (typically voice) trafc or call-signaling trafc because the incoming rates of these
trafc types do not tolerate packet loss and delay. Instead, the maximum rates of these
trafc types should be controlled at their origin by call admission control mechanisms
(which are discussed in Chapter 9, Call Admission Control [CAC]) so that excessive realtime trafc is not allowed onto the network.
Policers as Markers
If none of the congured actions for conform or exceed includes a drop function, the
operation of the policer becomes that of a conditional marker. This is why the terms marker
and policer often are used interchangeably, as they are in this chapter.
Re-marking packets with a policer should be done with care, keeping in mind the overall
policies of the network. Packets typically are marked as close to the source as technically
and administratively feasible (either at the source itself, if trusted, or at a network trust
boundary). In these locations, the trafc typically is marked by application: voice, call
signaling, video, high-priority data, and so on. This can be thought of as a vertical
separation of trafc.
A policer, however, has no direct knowledge of applications and marks trafc (if congured
to perform marking) based on the trafc rate. This can be thought of as a horizontal
separation of trafc. A policer could have indirect knowledge of an application by virtue of
the class where the policer is appliedin this conguration, the application already has
been separated from other trafc into its own horizontal class. The policer itself, however,
does not look at the class; it looks at only the trafc rate for the trafc owing through it.
Committed Access Rate

CAR is the oldest policing tool offered in the Cisco IOS Software and is included in this
chapter primarily for historical reasons. However, the Cisco 7300 platforms still use CAR
in certain congurations. RFC-compliant policers are available in newer releases of the
Cisco IOS Software, so CAR is not generally recommended for QoS deployments.
108
In addition to packet-drop capabilities, CAR can classify and mark packets using IPP,
DSCPs, and QoS group settings. Example 4-1 illustrates how to use CAR rate limiting on
applications such as web and FTP trafc, to ensure available capacity for other trafc.
Example 4-1 CAR Policing Example
Router# sh run
interface Hssi0/0/0
description 45Mbps to Router2
rate-limit input access-group 101 20000000 24000 32000
conform-action set-prec-transmit 2 exceed-action set-prec-transmit 0
conform-action set-prec-transmit 2 exceed-action drop
rate-limit input 8000000 16000 24000 conform-action set-prec-transmit 5
exceed-action drop
ip address 200.200.14.250 255.255.255.252
access-list 101 permit tcp any any eq www
access-list 102 permit tcp any any eq ftp
Access list 101 denes and matches web (WWW) trafc. The rst rate-limit statement
works on trafc matched by this access list:
conform-action set-prec-transmit 2 exceed-action set-prec-transmit 0
It limits the rate of web trafc to 20 Mbps, with a normal burst size of 24,000 bytes and an
excess burst size of 32,000 bytes. Trafc that conforms to the rate (less than 20 Mbps) is
marked with an IP Precedence of 2; trafc that exceeds the rate is marked with an IP
Precedence of 0. No trafc is dropped by this statement.
Access list 102 denes and matches FTP trafc. The second rate-limit statement works on
trafc matched by this access list:
conform-action set-prec-transmit 2 exceed-action drop
It limits the rate of FTP trafc to 10 Mbps, with a normal burst size of 24,000 bytes and an
excess burst size of 32,000 bytes. Trafc that conforms to the rate (less than 10 Mbps) is
marked with an IP Precedence of 2; trafc that exceeds the rate is dropped.
The third rate-limit statement works on remaining trafc:
rate-limit input 8000000 16000 24000 conform-action set-prec-transmit 5
exceed-action drop
It limits this trafc to 8 Mbps, with a normal burst size of 16,000 bytes and an excess burst
size of 24,000 bytes. Trafc that conforms to the rate (less than 8 Mbps) is marked with an
IP Precedence of 5; trafc that exceeds the rate is dropped.
Policers
109
Class-Based Policing
Conguring class-based policing using the MQC syntax is an easy way to activate policing
for only certain classes of trafc. With class-based policing, class denitions represent
application separation, and policing is performed only on the classes congured in the
policy map. This generally includes AF and BE classes in which the drop priority is
increased on out-of-contract trafc (for instance, re-marking from AF21 to AF22 or AF23)
or the trafc is dropped outright.
Trafc in classes that are not policed contends for bandwidth availability along with all
other trafc on the interface. When no congestion exists, all trafc is transmitted. When
congestion is experienced, packets can be dropped out of nonpoliced classes as well as
policed classes.
Class-based policing uses class maps and policy maps. A class map identies the trafc to
be policed (or the policer can be applied to class-default to police everything). The policy
map then details the CIR, Bc, and other policing parameters, including the conform and
exceed actions to be taken.
Class-based policing was introduced in Cisco IOS Software Release 12.1(5)T. Enhancements
were made in Cisco IOS Software Release 12.2(2)T. Class-based policing is supported in
the CEF path and in the fast- and process-switching paths. The policer can be specied at
the interface, subinterface, or even ATM/Frame Relay PVC levels. Class-based policing
works on both unicast and multicast packets.
Class-Based Policing Benets

Class-based policing is the currently recommended tool for policing. Its major advantages
over CAR are summarized here:
Class-based policing is compliant with DiffServ RFCs (CAR is not).
CAR does not exist within the MQC syntax. Therefore, its statistics cannot be tied
back to the policy statistics shown by the show policy interface command.
Class-based policing statistics are available in the CISCO-CLASS-BASEDQOS-MIB, offering enhanced network management and monitoring capability.
The granularity of classication for class-based policing is far superior to that

available for CAR. For example, NBAR can be used with class-based policing but
not with CAR.
Policing feature enhancements (such as percentage-based bandwidth specication

and hierarchical policing) are made only to the class-based policing features, not to
CAR.
110
Single-Rate Three-Color Marker/Policer

An improvement to the single-rate two-color marker/policer algorithm is based on
RFC 2697, which details the logic of a single-rate three-color marker.
The single-rate three-color marker/policer uses an algorithm with two token buckets. Any
unused tokens in the rst bucket are placed in a second token bucket to be used as credits
later for temporary bursts that might exceed the CIR. The allowance of tokens placed in this
second bucket is called the excess burst (Be), and this number of tokens is placed in the
bucket when Bc is full. When the Bc is not full, the second bucket contains the unused
tokens. The Be is the maximum number of bits that can exceed the burst size.
This two token bucket mechanism allows three possible trafc conditions to be identied
(hence the term three-color). Trafc can be identied as follows:
ConformTo the CIR

ExceedThe CIR within the excess burst allowance credits
ViolateBeyond both the CIR and any excess burst allowance credits
For these rate-based decisions, the following actions can be specied:
ConformOptionally re-mark and transmit

ExceedDrop or optionally re-mark and transmit
ViolateDrop or optionally re-mark and transmit
Policing is used not only to drop out-of-prole packets, but also to re-mark them, thus
indicating to downstream dropping mechanisms that they should be dropped ahead of
the in-prole packets.
The single-rate three-color marker uses the following denitions within the RFC:
CIRCommitted information rate, the policed rate
TeToken count of EBS, the instantaneous number of tokens left in the EBS bucket.
CBSCommitted burst size, the maximum size of the rst token bucket
EBSExcess burst size, the maximum size of the second token bucket
TcToken count of CBS, the instantaneous number of tokens left in the CBS bucket
(Do not confuse the term Tc here with the earlier use of Tc in the context of time
elapsed for policing or shaping trafc intervals.)
BByte size of offered packet
Figure 4-3 illustrates the logical ow of the single-rate three-color marker/policer (two
token bucket) algorithm.
The single-rate three-color policers tolerance of temporary bursts, shown in Figure 4-4,
results in fewer TCP retransmissions and, thus, more efcient bandwidth utilization.
Furthermore, it is a highly suitable tool for marking according to RFC 2597 AF classes,
which have three colors (or drop preferences) dened per class (AFx1, AFx2, and AFx3).
Policers
Figure 4-3
RFC 2697 Single-Rate Three-Color Policer Logic (Two Token Bucket Algorithm)
CIR
Overflow
EBS
CBS
B < Tc
No
B < Te
No
Packet of Size B
Yes
Yes
Exceed
Violate
Action
Action
Action
Two
Token Bucket
Policer
Time
Offered Traffic
RFC 2697 Single-Rate Three-Color Policer Effect on Trafc Flow (Two Token Bucket Policer)
Offered Traffic
Figure 4-4
Conform
Time
Temporary bursts (marked Y)
are permitted in excess of the CIR
only if unused token credits
(marked G) have been
accumulated.
111
112
Temporary bursts (shown shaded above the line in the graph on the right in Figure 4-4) are
permitted in excess of the CIR only if unused token credits (shown shaded below the line
in the graph) have been accumulated. Otherwise, this trafc is dropped.
Example 4-2 shows the conguration to police trafc in class-default to a CIR of 256 kbps,
with a Bc of 8000 bytes and a Be of 8000 bytes. Note that, for this policer, the CIR is
dened in bits per second, but Bc and Be are dened in bytes.
Example 4-2 Class Default Policing Example
Router# sh run
policy-map RFC2697-POLICER
class class-default
police cir 256000 bc 8000 be 8000
conform-action set-dscp-transmit af31
exceed-action set-dscp-transmit af32
violate-action set-dscp-transmit af33
Two-Rate Three-Color Marker/Policer

The single-rate three-color marker/policer was a signicant improvement for policers, in
that it made allowance for temporary trafc bursts (as long as the overall average transmitted rate was equal to or below the CIR). However, the variation in the number of accumulated
excess burst credits could cause a degree of unpredictability in trafc ows. To improve on
this, a two-rate three-color marker/policer was dened in RFC 2698. This policer addresses
the peak information rate (PIR), which is unpredictable in the RFC 2697 model. Furthermore, the two-rate three-color marker/policer allows for a sustainable excess burst (negating the need to accumulate credits to accommodate temporary bursts) and allows for
different actions for the trafc exceeding the different burst values.
The two-rate three-color marker/policer was introduced in Cisco IOS Software Release 12.2(4)T
and uses the following parameters to meter the trafc stream:
PIRPeak information rate, the maximum rate that trafc ever is allowed
PBSPeak burst size, the maximum size of the rst token bucket
CIRCommitted information rate, the policed rate
CBSCommitted burst size, the maximum size of the second token bucket
TpToken count of CBS, the instantaneous number of tokens left in the PBS bucket
TcToken count of EBS, the instantaneous number of tokens left in the CBS bucket
BByte size of offered packet
Policers
113
The two-rate three-color policer also uses an algorithm with two token buckets, but the
logic varies slightly. Instead of transferring unused tokens from one bucket to another, this
policer has two separate buckets that are lled each second with two separate token rates.
The rst bucket is lled with the PIR number of tokens and the second bucket is lled with
the CIR number of tokens. In this model, the Be works the same as the Bc, except for the
PBS bucket (not the CBS bucket). This means that Be represents the peak limit of trafc
that can be sent during a subsecond interval.
The logic varies further in that the initial check is to see whether the trafc is within the
PIR. Only then is the trafc compared against the CIR. (In other words, a violate condition
is checked for rst, then an exceed condition, and nally a conform condition, which is the
reverse of the logic of the previous model.) This logic is illustrated in Figure 4-5.
Figure 4-5
RFC 2698 Two-Rate Three-Color Policer Logic (Two Token Bucket Algorithm)
PIR
CIR
CBS
PBS
Packet of Size B
B > Tp
Yes
No
B > Tc
No
Yes
Violate
Exceed
Conform
Action
Action
Action
The two-rate three-color marker allows for sustainable excess bursts (and is not dependent
on accumulating credits) and has a hard-top peak limit, as shown in Figure 4-6.
114
Figure 4-6
RFC 2698 Two-Rate Three-Color Policer Effect on Trafc Flow (Two Token Bucket Policer)
Y
PIR
Offered Traffic
Offered Traffic
Two
Token Bucket
Policer
Time
CIR
Time
Sustained Excess Bursts (marked Y) are

permitted in excess of the CIR (no accumulation
of unused token credits is necessary) but only
until the PIR. Traffic above the PIR (circled)
is subject to the violate action.
Sustained excess bursts (shown shaded) are permitted in excess of the CIR (no
accumulation of unused token credits is necessary), but only until the PIR is reached.
Example 4-3 shows the conguration to police trafc on class-default to a CIR of 8000 bps,
a Bc of 1000 bytes, a Be of 2000 bytes, and a PIR of 10000 bps. Note that, for this policer,
CIR and PIR are dened in bits per second, but Bc and Be are dened in bytes.
Example 4-3 Two-Rate Three-Color Policer Example
Router# sh run
policy-map RFC2698-POLICER
class class-default
police cir 8000 bc 1000 pir 10000 be 2000
exceed-action set-dscp-transmit af32
violate-action set-dscp-transmit af32
Hierarchical Policing
It is advantageous to police some applications at multiple levels. For example, it might be
desirable to limit all TCP trafc to 10 Mbps, while at the same time limiting FTP trafc
(which is a subset of TCP trafc) to no more than 1.5 Mbps. To achieve this nested policing
requirement, hierarchical policing can be used. Two-level hierarchical policing was introduced
in Cisco IOS Software Release 12.1(5)T. Later, in Release 12.2.1(3)T, three-level hierarchical
policing was introduced for the 7200 and 7500 platforms.
The policer at the second level in the hierarchy acts on packets transmitted or marked by
the policer at the rst level. Therefore, the second level does not see any packets that the
Policers
115
rst level drops. The sum of packets that the lower-level policers see is equal to the sum of
packets that the higher-level policer transmits or marks. This feature supports up to three
nested levels.
Example 4-4 shows the conguration for the nested, two-level, hierarchical policing of TCP
and FTP trafc.
Example 4-4 Nested, Hierarchical Policing Example
Router# sh run
policy-map FTP-POLICER
class FTP
police cir 1500000
exceed-action drop
!
policy-map TCP-POLICER
class TCP
police cir 10000000
exceed-action drop
service-policy FTP-POLICER
Multiaction Policing
At times, packets need to be marked at both Layer 2 and Layer 3. To accommodate such a
requirement, a multiaction policer was introduced in Cisco IOS Software Release 12.2(8)T.
The commands in Example 4-5 illustrate the conguration.
Example 4-5 Multiaction Policer CLI
Router(config)# policy-map MULTIACTION-POLICER
Router(config-pmap)# class class-default
(config-pmap-c)#police cir [bc] [be] ?
conform-action action1
exceed-action
exceed-action
exceed-action
exceed-action
action1
action2
action3
action4
violate-action action1
violate-action action2
For each type of trafc, such as the conforming trafc, multiple actions can be specied.
The same is true for the exceeding trafc and the violating trafc. Example 4-6 illustrates
how to mark exceeding trafc with both IP Precedence 4 (at Layer 3) and the Frame Relay
DE bit (at Layer 2).
116
Example 4-6 Multiaction Policer Example

Router# sh run
policy-map MULTIACTION-POLICER
class class-default
police cir 10000 pir 2000000
exceed-action set-prec-transmit 4
exceed-action set-frde
violate-action set-prec-transmit 2
violate-action set-frde-transmit
Example 4-7 shows how a packet can be marked with DSCP AF31 (at Layer 3) and MPLS 3
(at Layer 2) at the same time.
Example 4-7 Multiaction Policer Example
Router# sh run
policy-map MULTIACTION-POLICER2
class class-default
police cir 10000
conform-action set-mpls-exp-topmost-transmit 3
Color-Aware Policing
RFC 2697 and RFC 2698 describe three-color policers, meaning that the packets can be
colored to three separate values to indicate whether they conform to, exceed, or violate the
policing conditions. The single-rate three-color marker and the two-rate three-color marker
initially were implemented (in Cisco IOS Software Releases 12.2[2]T and 12.2[4]T,
respectively) to operate in color-blind mode. This means that the policer assumes that the
packet stream previously was uncolored. The RFCs also dene a color-aware mode, which
means that the policer assumes that some preceding entity already has colored the packet
stream. At the time of this writing, the color-aware mode is available only in Cisco IOS
Software Release 12.0.26S; it is not yet available in any 12.2T release.
Table 4-2 shows the two-rate policer decisions for color-blind versus color-aware modes.
The number of bytes in the packet under consideration is indicated by pkt-length.
Percentage-Based Policing
Most networks contain a wide array of interfaces with different bandwidths. If absolute
bandwidth rates are used in policing policies, the policy must be re-entered for each different interface size. This reduces policy modularity and makes policy management more
cumbersome across the enterprise. It often is desirable to have an overall network policy in
which, for example, FTP trafc is not to exceed 10 percent of the bandwidth on any interfaceregardless of absolute speed. This can be achieved using percentages in the policing
statements. Thus, a single policy can be reused across many interfaces in the network.
Policers
Table 4-2
117
Color-Blind Versus Color-Aware Policing Modes for the Two-Rate Policer

Color Blind
Color Aware
If (pkt-length > PIR + PBS), mark the packet as

violating.
If the packet already is marked as violating

OR
(pkt-length > PIR + PBS),
mark the packet as violating.
If (pkt-length < PIR + PBS) AND (pkt-length >

CIR + CBS), mark the packet as exceeding.
If the packet already is marked as exceeding

OR
((pkt-length < PIR + PBS) AND (pkt-length >
CIR + CBS)),
mark the packet as exceeding.
Otherwise, mark the packet as conforming.
Otherwise, mark the packet as conforming.
The command for this is as follows:

police cir percent xx
police cir percent xx pir percent xx
Only the CIR and PIR values can be specied with percent, not the burst sizes; the burst
sizes are congured in units of milliseconds. If the CIR is congured in percent, the PIR
also must be. When the service-policy is attached to an interface, the CIR (and PIR, if
congured) is determined as a percentage of the interface bandwidth. If the interface
bandwidth is changed, the CIR and PIR values and burst sizes automatically are recalculated
using the new interface bandwidth value. For subinterfaces, the bandwidth of the main
interface is used for the calculation. For ATM, the VC bandwidth is used; for Frame Relay,
the CIR value of the PVC is used.
If the percent feature is used in a second- or third-level policy, the bandwidth of the lowerlevel policy statement is determined by the conguration of the higher or parent level.
Table 4-3 summarizes this decision process.
Defaults
The burst parameters and actions of the policer are optional. The default conform action
is to transmit the packet, and the default exceed action is to drop the packet. The default
violate option is the same as the exceed action. The default burst values are 250 ms of the
specied trafc rate.
118
Table 4-3
CIR/PIR Policing Behavior Based on Percentages

Configuration
Decision
CIR and PIR congured
If the conform action is drop, the bandwidth is 0. (This is most

likely not a practically useful conguration, but it is nevertheless the
result of the police statement if this should be congured.)
If the exceed action is drop, the CIR specication is used as the
bandwidth.
If the violate action is drop, the PIR specication is used as the
bandwidth.
If the conform action is drop, the bandwidth is 0. (This is most likely
not a practically useful conguration, but it is nevertheless the result
of the police statement if this should be congured.)
CIR congured
If the exceed action is drop, the CIR specication is used as the

bandwidth.
If the violate action is drop, the CIR specication is used as the
bandwidth.
Neither CIR nor PIR
congured
If this is a second- or third-level policy, the rate is determined by

the parent class.
If this is a rst level policy, the interface of the PVC bandwidth is
used.
Shapers
Similar to policers, shapers meter the transmission rate of packets through a network.
However, unlike policers, shapers delay (instead of drop or re-mark) packets that exceed
the CIR. Such delaying smoothes out bursts in trafc ows and allows for conformance to
SLAs, as shown in Figure 4-7.
Figure 4-7
Generic Trafc Shaping Effect on Trafc Flow

Without Traffic Shaping
Line
Rate
With Traffic Shaping

CIR
Traffic shaping limits the transmit rate of traffic

to a value (CIR) lower than the interfaces line rate.
Shapers
119
Shaping is crucial on nonbroadcast multiaccess (NBMA) topologies, such as ATM and

Frame Relay, in which potential speed mismatches exist.
Figure 4-8
NBMA Scenarios Requiring Trafc Shaping
Central
Site
T1
Frame Relay
or ATM
64
kbps
T1
T1
T1
CIR =
64 kbps
Remote Sites
In Figure 4-8, a trafc shaper is used in the following situations:
A line speed mismatchIn this case, the central site has a T1 link, but the remote
site has only a 64-kbps link. Without trafc shaping, frames and cells could
accumulate and drop in the carriers cloud.
Remote to central site aggregate oversubscriptionIn this case, if both remote

sites begin transmitting at the same time, multiple remote sites (each with T1 links)
begin sending at line rates back to a central site (which has only a single T1 link),
thereby oversubscribing the central T1 and again causing drops within the carriers
cloud.
120
SLA enforcementThe enterprise has contracted for a 64-kbps CIR from its carrier,
and the carrier offers no service guarantees for trafc exceeding this rate. Service
guarantees are critical when planning IP telephony or IP videoconferencing
deployments.
Because shaping involves buffering, various scheduling techniques can be used when the
shaping buffer starts lling. These scheduling techniques are discussed in more detail in
Chapter 5, Congestion-Management Tools.
Shaping Algorithms
Similar to policers, shapers use token bucket algorithms. By default, some shapers set the
Bc to equal CIR/8, which yields an interval (Tc) of 125 ms. Although this interval value
might be adequate for data applications, it introduces unnecessary interpacket delays for
real-time networks. Consider the example shown in Figure 4-9.
Figure 4-9
Shaping Behavior with Default Bc Values
Data Flow on a 128-kbps Line, Shaping to 64 kbps

Interval =
Bc
CIR
125 ms Interval =
8000 bits
64,000 bps
Cisco Default Bc = 1/8 CIR = 125-ms Interval

0
Bits Per Interval of

bits
Time at 128-kbps Rate
16,000
bits
32,000
bits
48,000
bits
64,000
bits
80,000
bits
96,000
bits
112,000 128,000
bits
bits
Line Rate
128 kbps
Net Result:
8000 * 8 = 64 kbps
62.5 ms
0 ms
125 ms 250 ms
When 8000 bits (Bc) transmitted

credits are exhausted and no more
packet flow in that interval
this happens at the 62.5-ms point
of the interval.
375 ms 500 ms 625 ms
750 ms 875 ms 1000 ms
When a new interval begins, Bc (8000 bits)

credits are restored and transmission
may resume. Pause in transmission is
every 62.5 ms in this case.
A shaper uses the CIR, Bc, and Be to smooth a trafc stream to a specied rate. It achieves
the given CIR by dividing the line speed of the interface into equal-length time slots
(intervals, or Tc). Then it sends a smaller portion (Bc) of the trafc during each time slot.
Shapers
121
The time slot size is governed by the Bc parameter (Tc = Bc/CIR). In Figure 4-9, a line rate
of 128 kbps is shaped to 64 kbps. For Frame Relay trafc shaping, the Bc, by default, is one
eighth of the CIR (or 8 kbps). Each second is divided into eight time slots of 125 ms each,
and the shaped rate of 64,000 bps is divided into eight bursts of 8000 bps each (Bc). Each
burst takes 62.5 ms to transmit (at a 128-kbps line rate), so the shaper transmits information
for the rst 62.5 ms of each 125-ms time slot and is silent for the remaining 62.5 ms of the
time slot. Over the span of a second, this achieves the average rate of CIR.
The Be value is used to determine the peak rate of sending and is calculated as follows:
Peak Rate = CIR (1+ Be / Bc)
Peak-rate shaping allows the router to burst higher than average-rate shaping. However,
when peak-rate shaping is enabled, any trafc that exceeds the CIR could be dropped if the
network becomes congested.
The design goal for one-way latency across a real-time network is 150 ms (per the G.114
specication for voice end-to-end delay), and the jitter target is less than 10 ms per hop.
Introducing up to 125 ms of shaping delays at a single hop destroys VoIP quality. Therefore,
it is recommended that shaped interfaces carrying real-time trafc be shaped to 10-ms
intervals (Tc). Because the Tc cannot be administered directly by Cisco IOS commands,
Bc tuning is required to affect Tc indirectly. This interval value can be achieved (remember,
Tc = Bc / CIR) by setting the Bc equal to CIR/100, which results in a Tc value of 10 ms.
The recommended value for Bc on an 64-kbps interface/PVC carrying real-time trafc is,
therefore, 64,000 / 100, which equals 640 bits and represents 10 ms of transmission on a
64-kbps line.
Shaping on ATM and Frame Relay Networks

ATM and Frame Relay are NBMA topologies that have inherent and specic requirements
for trafc shaping. Therefore, these topologies have customized shaping tools that are not
available to other topologies. These custom-shaping tools include ATM trafc contracts and
Frame Relay trafc shaping.
ATM Trafc Contracts

In ATM, shaping is an integral part of the PVC specication. This is because ATM achieves
different QoS levels by conguring different PVC trafc contracts, such as real-time
variable bit rate (VBR-RT), non-realtime variable bit rate (VBR-NRT), available bit rate
(ABR), and unspecied bit rate (UBR). The ATM trafc contract shaping commands are
summarized in Table 4-4.
122
Table 4-4
ATM Trafc Contract PVC Parameter Conguration Commands

Command
Purpose
Router(cong-if-atm-vc)# abr
output-pcr output-mcr
Congures the ABR
Router(cong-if-atm-vc)# ubr
output-pcr
Congures the UBR
Router(cong-if-atm-vc)# ubr+
output-pcr output-mcr
Congures the UBR with a minimum guaranteed rate
Router(cong-if-atm-vc)# vbr-nrt
output-pcr output-scr output-mbs
Congures the non-realtime VBR
Router(cong-if-atm-vc)# vbr-rt
peak-rate average-rate burst
Congures the real-time VBR
The -pcr and -mcr arguments are the peak cell rate and minimum cell rate, respectively. The
-scr and -mbs arguments are the sustainable cell rate and maximum burst size, respectively.
The parameters specied on the ATM PVC conguration determine the shaped average or
peak rates, as well as the burst parameters, as shown in Example 4-8.
Example 4-8 ATM PVC Conguration
Router# sh run
interface ATM3/0.1 point-to-point
description OC3 Link to Site-B
ip address 10.2.12.1 255.255.255.252
pvc 0/12
vbr-nrt 149760 149760
General Cisco IOS shaping tools, such as generic trafc shaping and class-based shaping,
are not supported on ATM interfaces, including digital subscriber line (DSL).
Frame Relay Trafc Shaping

Trafc shaping is imperative for Frame Relay PVCs if QoS guarantees are required over
them. The initial shaping mechanism developed for Frame Relay interfaces was Frame
Relay trafc shaping (FRTS). Example 4-9 shows a sample conguration of FRTS.
Example 4-9 Frame Relay Trafc Shaping
Router# sh run
interface Serial0/1
no ip address
encapsulation frame-relay
frame-relay traffic-shaping
!
interface Serial0/1.50 point-to-point
Shapers
123
Example 4-9 Frame Relay Trafc Shaping (Continued)

description FR Link to BRANCH#50
bandwidth 1536
ip address 10.200.50.1 255.255.255.252
frame-relay interface-dlci 150
class FRTS-1536
!
map-class frame-relay FRTS-1536
frame-relay cir 1536000
frame-relay bc 15360
frame-relay be 0
frame-relay mincir 1536000
no frame-relay adaptive-shaping
In line with the previously discussed shaping recommendations, the Bc is set to CIR/100.
The Be is set to 0 to prevent any excess bursting. The problem with provisioning the Be in
real-time networks is that it can create buffering delays within a Frame Relay network
(because the receiving side can pull the trafc from a circuit only at the rate of Bc, not
Bc + Be). To remove the potential for buffering delays, the recommendation is to set the Be
to 0.
Frame Relay also has the capability to adapt to explicit congestion notices, either in the forward direction through forward explicit congestion notications (FECNs) or in the reverse
direction through backward explicit congestion notications (BECNs). Adaptive trafc
shaping even can be triggered by other notications, such as foresight or interface congestion (as of Cisco IOS Software Release 12.2[4]T). When adaptive shaping is enabled and
congestion notications have been received, the FRTS engine rates down the ow to the
minimum CIR (minCIR), which, by default, is CIR/2. Because such congestion adaptation
introduces variations in transmission rates and, thus, service levels, the recommendation is
to disable adaptive shaping in networks carrying real-time trafc.
Class-Based Frame Relay Trafc Shaping

In Cisco IOS Software Release 12.2(13)T, FRTS functionality was migrated into the MQC
syntax using class-based shaping commands. This is the currently recommended method
for conguring trafc shaping on Frame Relay data-link connection identiers (DLCIs).
Translating Example 4-9 into the class-based FRTS syntax yields the results shown in
Example 4-10.
Example 4-10 Frame Relay Trafc Shaping Using Class-Based Syntax
Router# sh run
policy-map CB-FRTS-1536
class class-default
shape average 1536000 15360 0
continues
124
Example 4-10 Frame Relay Trafc Shaping Using Class-Based Syntax (Continued)
!
...
!
interface Serial0/1
no ip address
!
bandwidth 1536
ip address 10.200.50.1 255.255.255.252
class FRTS-1536
!
...
!
service-policy output CB-FRTS-1536
The class-based FRTS conguration retains remnants of FRTS, such as the need to associate the DLCI to a Frame Relay map class and then attach the class-based FRTS policy to
this map class.
Frame Relay Voice-Adaptive Trafc Shaping

Supporting voice trafc with predictable QoS on a Frame Relay PVC requires trafc
shaping to be deployed on the PVC and shaped strictly to the CIR. This is because any
trafc in excess of the CIR could be marked as Frame Relay DE-bit 1 and could be dropped
by the Frame Relay backbone network during congestion (because it violates the SLA).
Many customer edge networks, however, are designed and congured to oversubscribe the
CIR because the backbone network seldom exhibits congestion and, if it does, the higherlayer protocols take care of recovering the session. Voice trafc cannot be treated in the
same manner because the QoS for voice needs to be predictable and guaranteed.
The Frame Relay backbone network does not know which frames contain voice and which
contain data. Therefore, if any trafc violates the CIR, the backbone switch does not know
how to drop only data and not to drop voiceeven if the voice trafc by itself is below the
contracted SLA rate. This necessitates shaping strictly to the CIR to ensure that voice trafc
is protected. Adaptive shaping (through BECNs, FECNs, and other notication mechanisms)
cannot resolve this problem because by the time the shaper receives the congestion notices
and reacts to them, many voice packets might already have been delayed or dropped.
Therefore, to address the needs of voice trafc in a Frame Relay network, the Frame Relay
voice-adaptive trafc shaping (FR-VATS) feature was introduced in Cisco IOS Software
Release 12.2(15)T.
Shapers
125
FR-VATS monitors the Frame Relay PVC. When voice activity is present, it automatically
shapes to the CIR (which, in this specic case, is congured as the minCIR). However, if
no voice is present, it allows the trafc to burst to the line rate (which, in this specic case,
is congured as the CIR). This unique feature determines the presence of voice based on
packets entering a priority class or low-latency queue (a scheduling feature that is discussed
in Chapter 5). It then automatically determines whether to shape the trafc to minCIR
(voice present) or not (no voice present). Similarly, FR-VATS automatically engages Frame
Relay fragmentation (FRF.12) on links less than 768 kbps when voice packets are present,
to prevent unnecessary serialization delays. (FRF.12 is discussed in more detail in Chapter 7,
Link-Specic Tools.)
Because FR-VATS automatically detects the presence of voice, there could be some brief
quality degradation in the rst couple of seconds of the rst voice call made across a PVC.
This occurs while the interfaces comprising the PVC recongure themselves to shape to the
minCIR and empty out their buffers. FR-VATS is not a predictive algorithm. The change in
behavior is triggered by voice packets owing across the PVC.
The FR-VATS deactivation period is tunable and, by default, is set to 30 seconds. If tuned,
this timer should be set so that the feature will not turn off frequently during normal
business use (for example, between every two calls). The feature works better on PVCs that
always have at least one voice call present during daytime use and relinquish shaping only
at night.
Example 4-11 shows a conguration of FR-VATS.
Example 4-11 FR-VATS Conguration
Router# sh run
policy-map FR-VATS-768
class class-default
sets CIR to 768 kbps and Bc to minCIR/100
shape adaptive 364800
sets the minCIR to 364.8kbps
shape fr-voice-adapt deactivation 30 enables default FR VATS deactivation timer
!
...
!
interface Serial0/1
no ip address
serial restart_delay 0
frame-relay fragmentation voice-adaptive deactivation 30 enables FR Voice
Adaptive Fragmentation (sub-component of FR VATS)
!
bandwidth 768
ip address 10.200.50.1 255.255.255.252
class FRTS-768
applies the FR map-class to the DLCI
continues
126
Example 4-11 FR-VATS Conguration (Continued)

!
...
!
service-policy output FR-VATS-768
frame-relay fragment 480
!
sets fragment size for 384 kbps (minCIR) link
Generic Trafc Shaping

Generic trafc shaping (GTS) is the oldest shaping tool in the Cisco IOS Software and is
included in this chapter for historical reasons only. It was introduced in Cisco IOS Software
Release 11.2 and is enabled using the trafc-shape rate command. This feature is not
recommended for use in QoS deployments because it lacks many of the RFC-compliant
features contained in the class-based shaping tools discussed in this chapter. Instead, classbased shaping should be used wherever possible. In older IOS releases for Frame Relay
interfaces, FRTS should be used.
Class-Based Shaping
Class-based shaping (available within the MQC syntax) uses class maps and policy maps.
A class map identies the trafc to be shaped (or the shaper can be applied to class-default
to shape everything). The policy map then details the CIR, Bc, Be, and other shaping
parameters.
Class-based shaping was introduced in Cisco IOS Software Release 12.1(2)T.
Conceptually, it was a migration of GTS into the MQC syntax. However, the class-based
shaping code is different from GTS. This feature should not be assumed to behave in
exactly the same manner as GTS.
The command to congure class-based shaping follows:
shape [average | peak] [cir] [burst_size [excess_burst_size]]
Example 4-12 uses class-based shaping on a T1 interface to shape to 768-kbps CIR with
the Bc set to the recommended value for real-time networks (CIR/100 and Be set to 0).
Example 4-12 Class-Based Shaping on a T1 Interface
Router# sh run
policy-map CBS-768
class class-default
Shapers
127
Hierarchical Class-Based Shaping

As with hierarchical policing, sometimes shaping policies need to shape aggregate levels
and provide additional QoS functions on sublevels of trafc. Hierarchical shaping was
introduced in Cisco IOS Software Release 12.1(2)T.
For example, consider the case of a service provider that wants to shape a customers aggregate trafc to 384,000 and, within that, wants to provide a bandwidth guarantee of 200 kbps
for transactional data and a bandwidth guarantee of 100 kbps for bulk data. This can be
achieved with a nested, or hierarchical, shaping policy, as shown in Example 4-13.
Example 4-13 Hierarchical Class-Based Shaping
Router# sh run
policy-map CUSTOMER1-SHAPING-AND-QUEUING-POLICY
class CUSTOMER1-TRAFFIC
shape average 384000
service-policy CUSTOMER1-QUEUING-POLICIES
!
policy-map CUSTOMER1-QUEUING-POLICIES
class CUSTOMER1-TRANSACTIONAL-DATA
bandwidth 200
class CUSTOMER1-BULK-DATA
bandwidth 100
!
parent policy
includes the child policy

child policy
Percentage-Based Shaping
As with policing, shaping can be stated as a percentage of bandwidth instead of absolute
bandwidth. This feature enhancement was introduced in Cisco IOS Software Release 12.2(13)T
and is enabled with the following command:
shape [average | peak] percent xx
Burst sizes cannot be congured using percentages, but they can be specied in terms of
milliseconds on the command. When a percentage conguration is used for average or peak
shaping rates, the default burst sizes automatically are assigned and can be seen using the
show policy interface command, as shown in Example 4-14. When the service policy is
attached to an interface, the CIR value is determined from the interface bandwidth in
conjunction with the congured CIR percent value.
Example 4-14 Percent-Based Shaping
Router#show policy-map interface fastEthernet 0/1 output
FastEthernet0/1
Service-policy output: ex-shape
Class-map: example (match-all)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: access-group 101
continues
128
Example 4-14 Percent-Based Shaping (Continued)

Traffic Shaping
Target/Average
Byte
Sustain
Rate
Limit bits/int
80 (%)
0 (ms)
80000000/80000000 500000 2000000
Adapt Queue
Packets
Bytes
Active Depth
0
0
0
Class-map: class-default (match-any)
15 packets, 1225 bytes
5 minute offered rate 0 bps, drop rate
Match: any
Excess
Interval
bits/int (ms)
0 (ms)
2000000
25
Packets
Bytes
Delayed
Delayed
0
0
Increment
(bytes)
250000
Shaping
Active
no
0 bps
Distributed Trafc Shaping

On distributed platforms, such as the Cisco 7500 VIP, class-based shaping is not supported.
Instead, a counterpart feature, distributed trafc shaping (DTS), is used.
Although DTS on the Cisco 7500 is very similar to class-based shaping on nondistributed
platforms, there are two main differences to keep in mind. The rst is that, with DTS, the
CIR must be dened in multiples of 8000. The second is that the Cisco 7500 VIP requires
the interval (Tc) to be dened in increments of 4 ms. Because the target interval for all
platforms is 10 ms, which is not evenly divisible by 4 ms, the recommendation for Cisco
7500 VIP is to use an interval of 8 ms. The interval can be set to 8 ms by dening the burst
using the following formula:
Bc = CIR / 125
Detailed examples of DTS in WAN environments are provided in Chapter 13, WAN
Aggregator QoS Design. The behavior of trafc shaping for the 7500 platforms currently
is being reworked, but the previous functionality exists in Cisco IOS Software releases up
to 12.3 mainline and 12.3T.
Further Reading
DiffServ Policing Standards

Further Reading
129
Policing
Trafc policing (12.1.5T): http://www.cisco.com/univercd/cc/td/doc/product/

software/ios121/121newft/121t/121t5/dtpoli.htm.
Trafc policing enhancements (12.2.2T): http://www.cisco.com/univercd/cc/td/

doc/product/software/ios122/122newft/122t/122t2/ftpoli.htm.
Two-rate policer (12.2.4T): http://www.cisco.com/univercd/cc/td/doc/product/

software/ios122/122newft/122t/122t4/ft2rtplc.htm.
Policer enhancementmultiple actions (12.2.8T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios122/122newft/122t/122t8/ftpolenh.htm.
Percentage-based policing and shaping (12.2.13T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios122/122newft/122t/122t13/ftpctpol.htm.
Color-aware policer (12.0.26S): http://www.cisco.com/univercd/cc/td/doc/product/

software/ios120/120newft/120limit/120s/120s26/12s_cap.htm.
ATM PVC Trafc Parameters
Conguring ATM trafc parameters: http://www.cisco.com/univercd/cc/td/doc/

product/software/ios122/122cgcr/fwan_c/wcfatm.htm#1001126.
Frame Relay Trafc Shaping
Adaptive Frame Relay trafc shaping for interface congestion (12.2.4T): http://
ft_afrts.htm.
MQC-based Frame Relay trafc shaping (12.2.13T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios122/122newft/122t/122t13/frqosmqc.htm.
Percentage-based policing and shaping (12.2.13T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios122/122newft/122t/122t13/ftpctpol.htm.
Frame Relay queuing and fragmentation at the interface (12.2.13T):

122t/122t13/frfrintq.htm.
Frame Relay voice-adaptive trafc shaping and fragmentation (12.2.15T):

122t/122t15/ft_vats.htm.
130
Trafc Shaping
Generic trafc shaping (11.2): http://www.cisco.com/univercd/cc/td/doc/product/

software/ios112/112cg_cr/1rbook/1rsysmgt.htm#23613.
Class-based shaping (12.1.2T): http://www.cisco.com/univercd/cc/td/doc/product/

software/ios121/121newft/121t/121t2/clsbsshp.htm.
Distributed trafc shaping (12.1.5T): http://www.cisco.com/univercd/cc/td/doc/

product/software/ios121/121newft/121t/121t5/dtdts.htm.
Scheduling and queuing

Legacy Layer 3 mechanisms
Currently recommended Layer 3 mechanisms
Layer 2 queuing tools
Tx-ring
PAK_priority
CHAPTER
Of all the mechanisms within the QoS toolset, congestion-management tools provide the
most signicant impact on application service levels. Congestion-management tools, also
known as queuing tools, apply to interfaces that may experience congestion. This could be
the case in either a WAN or a LAN environment (although LAN queuing tools are discussed
in Chapter 10, Catalyst QoS Tools) because speed mismatches can occur in either setting.
Whenever packets enter a device faster than they can exit it, the potential for congestion
exists and queuing tools apply.
It is important to recognize that queuing tools are activated only when congestion exists
on an interface. When no congestion exists on an interface, packets are sent out over the
interface as soon as they arrive. However, when congestion occurs, packets must be buffered,
or queued, to mitigate dropping.
Packet markings at either Layer 2 or Layer 3, or the absence of a marking on a packet,
inuence queuing policies so that the router can reorder packets before transmission.
Therefore, queuing policies are usually complementary and depend on classication and
marking policies within converged networks.
This chapter examines the difference between scheduling and queuing, and the different
OSI layers where queuing might occur. Legacy Layer 3 queuing algorithms are reviewed to
provide context for the more recent queuing algorithms, such as Class-Based Weighted Fair
Queuing (CBWFQ) and Low-Latency Queuing (LLQ), that evolved from them. LLQ and
CBWFQ are examined in detail because these are the principal recommended queuing
mechanisms for converged networks in which different trafc types such as voice, video,
and data all share the same transmission media.
Lower-level queuing algorithms, such as Layer 2 mechanisms and the Tx-ring, also are
discussed because these tools have an impact on successful end-to-end QoS design. Finally,
the internal Cisco IOS mechanism for protecting routing and control packets (PAK_priority)
within the router is introduced.
134
Chapter 5: Congestion-Management Tools

The phrase scheduling tools refers to the set of features that determines how a frame, cell,
or packet exits a device. Whenever packets enter a device faster than they can exit it, as is
the case with interface speed mismatches (for example, Fast Ethernet trafc heading to a
serial WAN interface), a point of potential congestion, or a bottleneck, occurs. Devices have
buffers that allow for temporarily storing and subsequent scheduling of these backed-up
packets. This process commonly is referred to as queuing. As these buffers (or queues) ll,
packets can be reordered so that higher-priority packets exit the device sooner than lowerpriority ones. Figure 5-1 shows a general example of queuing.
Figure 5-1
Generic Queuing Example

Queuing
Scheduling
Inbound Packets
Outbound
Packets
Packets are assigned

various queues when the
output interface is congested
Conceptually, queuing and scheduling are complementary but intertwined processes. These
terms quite often are incorrectly used interchangeably:
Queuing is the logic of ordering packets in linked output buffers. It is important to

recognize that queuing processes are engaged only when the interface is experiencing
congestion and are deactivated shortly after the interface congestion clears.
Scheduling is the process of deciding which packet to transmit next. Scheduling

occurs when the interface is experiencing congestion, but also (unlike queuing) when
the interface is not experiencing congestion (that is, there is still a decision, albeit a
simple one, of which packet should be transmitted next, even if there is no congestion
on the interface). If there is no congestion on the interface, packets are transmitted as
they arrive. If the interface is experiencing congestion, queuing algorithms are engaged.
However, the scheduler still has to decide which queue to service next. This decision
is made using various types of scheduling logic algorithms:
Strict priorityThe lower-priority queues are served only if the higherpriority queues are completely empty. This type of scheduling exhibits the
potential to starve trafc in lower-priority queues.
135
Round-robinQueues are served in a sequence. Although they never

starve trafc, round-robin schedulers exhibit the potential to cause unpredictable delay in queues that hold real-time, delay-sensitive trafc that
requires a more strict priority-scheduling algorithm.
Weighted fairPackets in the queues are weighted, typically by IP
precedence, so that some queues are served more frequently than others.
Although this solves the downsides of both the strict priority and roundrobin algorithms, it does not provide the bandwidth guarantee that real-time
ows require. The resulting bandwidth per ow instantaneously varies
based on the number of ows present and the weights of each of the other
ows.
The buffer space (or memory) for queues is nite in capacity and acts very much like a
funnel for water that is being poured into a small opening. If water continually is entering
the funnel much faster than it exits, eventually the funnel begins to overow from the top.
When queuing buffers begin overowing, packets might be dropped either as they arrive
(tail drop) or selectively, before all buffers are lled. Selective dropping of packets when
the queues are lling is referred to as congestion avoidance. Congestion-avoidance
mechanisms work best with TCP-based applications because selective dropping of packets
causes the TCP windowing mechanisms to throttle back and adjust the rate of ows to
manageable rates.
Congestion-avoidance mechanisms are complementary to queuing algorithms and are
discussed in more detail in Chapter 6, Congestion-Avoidance Tools. The relationship
between congestion-management tools (or scheduling algorithms) and congestion-avoidance
tools (or selective-dropping algorithms) is as follows: With congestion management, the
scheduling algorithms manage the front of a queue; with congestion avoidance, the
mechanisms manage the tail of a queue.
Congestion-management and congestion-avoidance tools employed in converged networks
include these:
Low-Latency Queuing (LLQ)Congestion management

Class-Based Weighted Fair Queuing (CBWFQ)Congestion management
Weighted Random Early Detection (WRED)Congestion avoidance
LLQ and CBWFQ are discussed in detail in this chapter. WRED is covered in Chapter 6.
To understand queuing, you rst must understand the various levels of queuing and how
queuing has evolved. A fundamental concept in Cisco IOS queuing is that queuing is
performed at multiple layers of the OSI model.
Layer 3 queuing subsystems operate at the network layer and are applied to Layer 3 packets
at a logical level. Because Cisco IOS Layer 3 queuing subsystems are independent of the
egress interface type, they can be applied to Asynchronous Transfer Mode (ATM), Frame
Relay, High-Level Data Link Control (HDLC), Point-to-Point Protocol (PPP), Multilink
136
Point-to-Point Protocol (MLP), or tunnel interfaces. It is important to recognize that Layer 3

queuing typically considers only the IP overhead in its bandwidth provisioning. Layer 2
overhead such as the ATM cell tax is not factored into the equation and, therefore, must be
provisioned explicitly for drop-sensitive applications such as Voice over IP (VoIP).
Layer 2 queuing subsystems accommodate media-specic requirements and idiosyncrasies.
For example, ATM and Frame Relay media support multiple Layer 2 logical interfaces
(for example, PVCs and DLCIs) per physical interface and require special scheduling to
accommodate such circuits. Layer 2 media-specic shapers, such as Frame Relay trafc
shaping, also interact with queuing, in that the shapers determine the output rate of packets
onto the interface and, therefore, push packets arriving too fast back into the queues.
When the Layer 2 queues ll up, they, in turn, push back packets into the Layer 3 queues.
Fragmentation and interleaving also are performed within Layer 2 queuing subsystems.
A nal queue, usually referred to as a transmit ring (or Tx-ring), is located within the device
driver itself. Tx-rings are media and hardware dependent and can operate quite differently
on different routers, cards, and modules. The purpose of the Tx-ring is to ensure that there
is always a packet ready to be placed onto the media, with the objective of driving interface
utilization to 100 percent. Furthermore, the Tx-ring indicates to the queuing algorithms
whether the interface is congested. (If the Tx-ring is full, the interface is congested.) When
the Tx-ring limit is reached, queuing algorithms engage. Some interfaces report the length
of the Tx-ring in packets. Other interfaces, such as ATM, report the length in particles.
(Particle length is variable but is set to 576 bytes for most ATM port adaptors.)
It is important to note that not all three levels of queuing are always present.
Legacy Layer 3 Queuing Mechanisms

There is a long history of queuing algorithms in Cisco IOS Software, not all of which are
covered in this chapter because they are not applicable to QoS deployment for converged
networks. Newer queuing and scheduling algorithms are simply combinations and enhancements of older queuing algorithms. For a historical perspective and an understanding of why
these mechanisms are insufcient for todays converged networks, it is helpful to review
some of these legacy queuing techniques. The three basic legacy queuing types (beyond
simple FIFO) are these:
Priority queuing (PQ)

Custom queuing (CQ)
Weighted Fair Queuing (WFQ)
Legacy Layer 3 Queuing Mechanisms
137
From these three basic algorithms, combinations and enhancements were developed. These
newer queuing mechanisms sought to combine the best features of the basic algorithms and,
at the same time, to minimize their drawbacks. These enhanced queuing algorithms include
the following:
IP RTP priority queuing (PQ-WFQ). This was a transient method and soon was
superceded by low-latency queuing.
CBWFQ
LLQ
Table 5-1 compares the attributes of the different algorithms and their suitability to
handling real-time, delay-sensitive trafc, such as voice. This trafc requires two things
from a queuing/scheduling algorithm:
An absolute bandwidth guarantee

A delay guarantee
Priority Queuing
The oldest queuing algorithm is PQ, which consists of only four queues (high, medium,
normal/default, and low). The scheduler begins by emptying the high queue and begins to
service lower queues only when upper queues are completely empty. This basic algorithm
proved very successful in handling real-time trafc, but it posed starvation possibilities to
lower-priority trafc ows.
Custom Queuing
CQ addressed the concern of trafc starvation, common in PQ scenarios. By introducing a
round-robin scheduler that was based on byte counts, not only did custom queuing prevent
bandwidth starvation, but it also became the rst queuing mechanism to provide bandwidth
guarantees. Although custom queuing (which supports up to 16 distinct queues) introduced
a fairness that PQ did not have, it lost the capability to provide strict priority to real-time
ows.
Weighted Fair Queuing

WFQ was developed to expand on the principle of fairness that CQ broached.
The fair-queuing algorithm within WFQ simply divides the interfaces bandwidth by the
number of ows, ensuring an equitable distribution of bandwidth for all applications.
138
Table 5-1 Queuing Algorithm Compariso

PQ
CQ
WFQ
PQ-WFQ
CBWFQ
LLQ
Per interface
Per protocol,
per interface
Per protocol,
per interface
IP prec, RSVP,
RTP Reserve,
protocol, port
RTP port for

PQ; IP prec
for WFQ
Class-based
classication
Class-based
classication
16
Per ow
1 PQ + WFQ Up to 256
classes
FIFO
Strict priority
Round robin
Weighted fair
(based on IP prec)
PQ: Strict
Weighted fair
(based on
bandwidth)
PQ: Strict
No
Yes for trafc in

highest-priority
queue only
No
No
Yes for PQ
trafc
No
Yes for PQ
trafc
Bandwidth
guarantee
No
No
Yes
No
Yes for PQ
trafc
Yes
Yes
Recommended
for voice
No
No
No
No
No
No
Yes
Classification
Number of
queues
Scheduling
WFQ: Weighted
fair (based on
IP prec)
1 PQ +
CBWFQ
CBWFQ: Weighted
fair (based on
bandwidth)
Delay guarantee
FIFO
Currently Recommended Layer 3 Queuing Mechanisms
139
By adding a xed weight (based on the packets IPP value) to the calculation, WFQ
introduced a method for moderately skewing bandwidth allotments by favoring higherpriority ows (priority being a function of IPP marking). Although such skewing preferred
certain ows over others, WFQ lost CQs capability to provide bandwidth guarantees
because bandwidth allocations continuously changed as ows were added or ended.
IP RTP Priority Queuing

IP RTP priority (PQ-WFQ) was a transient feature in the step-by-step evolution of
developing QoS mechanisms for voice.
IP RTP priority queuing was introduced between WFQ and LLQ in Cisco IOS Software
Release 12.0(5)T for interfaces and MLP links, and in Cisco IOS Software Release
12.0(7)T for Frame Relay. LLQ was introduced soon afterward in Cisco IOS Software
Release 12.0(7)T for interfaces and MLPPP links, and in Cisco IOS Software Release
12.1(2)T for Frame Relay. Therefore, the only legitimate current uses for IP RTP priority
queuing are on routers that handle voice trafc and that are running code levels between
these two releases. LLQ should be used in all other cases.
Currently Recommended Layer 3 Queuing

Mechanisms
Each of the previously described Layer 3 queuing tools had serious shortcomings in
handling the trafc mix required by converged networks. Either they handled data fairly
well or could prioritize real-time applications, such as voice. But none of these tools was
capable of handling both types of trafc efciently. Therefore, enhanced algorithms were
developed to leverage the strengths of each type of legacy algorithm and, at the same time,
to minimize their weaknesses.
This section provides an overview of the hybrid queuing mechanisms used today for Layer 3:
CBWFQ and LLQ.
Class-Based Weighted Fair Queuing

CBWFQ, introduced in Cisco IOS Software Release 12.0(5)T, is a hybrid queuing algorithm that combines the capability to guarantee bandwidth (from CQ) with the capability
to dynamically ensure fairness to other ows within a class of trafc (from WFQ).
CBWFQ enables the creation of up to 256 classes of trafc, each with its own reserved
queue. Each queue is serviced based on the bandwidth assigned to each class.
140
A major difference between WFQ and CBWFQ is that, with WFQ, the bandwidth of the
ow is calculated instantaneously; but with CBWFQ, a minimum bandwidth explicitly is
dened and enforced. Additionally, CBWFQ uses Modular QoS CLI (MQC)based class
maps for classication, which provide the richest and most granular recognition criteria
within the Cisco IOS QoS toolset. When class maps are combined with CBWFQ bandwidth
statements, minimum-bandwidth guarantees can be provisioned for almost any application.
Example 5-1 shows two applications of CBWFQ. First, a minimum bandwidth guarantee
is given to the MISSION-CRITICAL-DATA class. Second, all other trafc (which falls into
class-default) is fair queued.
Example 5-1 CBWFQ Policy for Multiple Levels of Data
policy-map WAN-EDGE
class MISSION-CRITICAL-DATA
bandwidth percent 20 sets the bandwidth for this class to 20%
class class-default
fair-queue sets WFQ as the queuing algorithm for unclassified traffic
NOTE
The percent keyword was added in Cisco IOS Software Release 12.0(7)T, allowing
bandwidth to be allocated as relative percentages of the link, or shaping rate, instead of
in absolute values (in kilobits per second). The percent keyword increases CBWFQ policy
modularity because the same policies can be applied to many interfaces, regardless of their
link speeds. Additionally, the percent keyword makes future expansion easier to manage
because the relative bandwidth allocations automatically will increase as the link speeds
themselves increase.
CBWFQ is a highly efcient algorithm for data applications, but it lacks the capability to
provide strict-priority servicing to real-time applications, such as voice or interactive video.
This is because it provides a bandwidth guarantee but not a latency guarantee. Therefore,
to service such real-time applications, a strict-priority queue was added to the CBWFQ
algorithm; the resulting algorithm was named low-latency queuing.
Low-Latency Queuing
LLQ is an enhanced combination of all three legacy queuing algorithms: PQ, CQ, and
WFQ. It was introduced in Cisco IOS Software Release 12.0(7)T.
LLQ is essentially CBWFQ combined with a strict PQ. Trafc assigned to the strict priority
queue, using the priority command, is serviced up to its assigned bandwidth before all
other CBWFQ queues are serviced.
NOTE
141
The original name for the LLQ algorithm was PQ-CBWFQ. Although this name was
technically correct, it was obviously clumsy from a marketing perspective. Hence, the
algorithm was renamed LLQ.
LLQ also has an implicitly dened burst parameter (added in Cisco IOS Software Release
12.1[3]T) that accommodates temporary bursts of real-time trafc. By default, LLQs are
provisioned to protect bursts up to 200 ms, dened in bytes. The burst parameter becomes
signicant in certain interactive video scenarios (which are discussed in more detail in
Chapter 13, WAN Aggregator QoS Design).
In Example 5-2, up to one-third of the links (or shapers) bandwidth is assigned for strict
priority treatment of voice.
Example 5-2 LLQ for VoIP and Multiple Levels of Data
policy-map WAN-EDGE
class VOICE
priority percent 33 sets the LLQ bandwidth to 33%
bandwidth percent 2
bandwidth percent 20
class class-default
fair-queue
NOTE
The percent keyword was added to the LLQ priority command in Cisco IOS Software
Release 12.2(2)T.
The next sections discuss further details about LLQ, such as its operation, the implicit
policing done by the queuing algorithm, bandwidth provisioning, and the lesser-known
interactions between LLQ, IPSec, cRTP, and LFI, and bundling technologies such as ATM,
MLP, and Frame Relay bundles.
LLQ Operation
LLQ is designed exclusively for converged networks that carry voice and interactive video.
It contains two components, a PQ for the real-time sensitive trafc and a class-based
complex of queues (CBWFQ). The PQ is the optimal queueing mechanism for real-time
trafc, whereas CBWFQ is the best queuing algorithm for data applications. Converged
networks require LLQ to be congured so that voice, interactive video, and data all are
given the service levels they require.
142
The Layer 3 queuing subsystem for LLQ is shown on the left side of Figure 5-2. In this
illustration, voice trafc is classied and placed in the PQ. When trafc is present in the PQ,
it is serviced with exhaustive priority treatment up to the bandwidth maximum congured
for the priority class (that is, voice is sent out rst until no more voice is present or the
bandwidth assigned is exceeded). The L2 subsystem is discussed later in this chapter.
Figure 5-2
LLQ Operation Logic
Link
Fragmentation
and Interleaving
LLQ/CBWFQ
Police
VoIP
IP/VC
PQ
Interleave
Signaling
TxRing
Packets
Out
Critical
Packets
In
Bulk
WFQ
CBWFQ
Fragment
Mgmt
Default
Layer 3
Queuing Subsystem
Layer 2
Queuing Subsystem
Device
Driver
LLQ Policing
The threat posed by any strict priority-scheduling algorithm is that it could starve lowerpriority trafc. To prevent this, the LLQ mechanism has a built-in policer. This policer (like
the queuing algorithm itself) engages only when the interface is experiencing congestion.
Therefore, it is important to provision the priority classes properly. If the total amount of
priority class bandwidth provisioned is lower than the total amount of voice trafc offered
to the PQ (including the Layer 2 overhead), the excess voice trafc might be dropped (by
the LLQ policer). As with any other policer, LLQs policer, when engaged, drops trafc
indiscriminately and affects the quality of all voice calls active on the interface (not just the
last one to go through). Strategies to accommodate this behavior of LLQ are discussed in
more detail in Chapter 9, Call Admission Control (CAC). In earlier Cisco IOS Releases,
the PQ of LLQ policed strictly to the bandwidth assigned, regardless of trafc present or
143
absent in other classes. In later releases, this operation changed to police strictly only if the
interface was congestedthat is if other classes were underutilized, PQ trafc would be
allowed to exceed its bandwidth without dropping packets.
Another benet provided by the implicit policer within the LLQ mechanism is the capability to use TDM for the single strict-priority queue. TDM abstracts the fact that only a single
strict-priority queue exists and allows for the conguration and servicing of multiple lowlatency queues.
In Example 5-3, separate priority classes on a dual-T1 MLP link are assigned for voice and
videoconferencing. Each priority class is policed (to 540 kbps for voice and 460 kbps for
video). Under the hood, however, there is only a single 1-Mbps PQ (540 kbps + 460 kbps),
which is time shared between these two applications by the implicit policer.
Example 5-3 LLQ Policy for VoIP, Interactive Video, and Multiple Levels of Data
policy-map WAN-EDGE
class VOICE
priority 540 creates a LLQ of 540 kbps
class INTERACTIVE-VIDEO
priority 460 creates a "second" LLQ of 460 kbps
bandwidth percent 2
class class-default
fair-queue
Bandwidth Provisioning
Most applications default to the best-effort default class. Because enterprises have hundreds,
if not thousands, of applications on their networks, it is a general best practice to provision
at least 25 percent of a links bandwidth for class-default. This recommendation has been
coded into older Cisco IOS Software releases for the lower-end platforms (Cisco 7200 and
below). If present, this restriction can be overridden manually (using the max-reservedbandwidth interface command).
A similar consideration must be given to the sum of all trafc assigned for all priority
classes. It is important to remember the goal of convergence: to allow voice, video, and data
to coexist transparently on a single network. If real-time applications dominate a network,
as in the case of a dominantly provisioned LLQ (in which the priority classes have the lions
share of the bandwidth), data applications might uctuate signicantly in their network
response times when voice/videoconference calls are made. This destroys the transparency
of the converged network. User feedback consistently has reected that most users prefer
consistent (even if moderately slower) application response times over varying response
times (as would occur if LLQs dominated WAN links). Cisco Technical Marketing testing
144
has revealed a conservative rule of thumb for priority class provisioning, which is to limit
the sum of all priority class trafc to no more than 33 percent of the WAN links capacity.
This 33 percent LLQ rule has been deployed widely and successfully.
Figure 5-3 illustrates these bandwidth-provisioning recommendationsnamely, that the
aggregate bandwidth of all priority classes should be no more than 33 percent of the links
capacity and that all bandwidth guarantees within LLQ should be no more than 75 percent
of the link capacity.
Figure 5-3
LLQ Bandwidth Provisioning
Voice
Call
Interactive
Signaling
Video
Critical
Data
Best-Effort
Data
LLQs 33%
Sum of all BW guarantees 75%
Link Capacity
Because of the policing function within LLQ, it is important to accurately provision

bandwidth required for voice calls. LLQ is a Layer 3 queuing mechanism and, as such,
accounts for only Layer 3 packet bandwidth. (It does not take Layer 2 headers into account.)
Furthermore, the Layer 3 bandwidth required by a voice call is determined by a number of
factors. These factors include the codec used, the sampling size, and the compression
techniques, such as cRTP (also known as RTP header compression). Cisco has provided a
voice bandwidth calculator to determine the correct bandwidth to provision in LLQ for
voice calls. The calculator is located on the web at http://tools.cisco.com/Support/VBC/jsp/
Codec_Calc1.jsp.
When the percentage-remaining (bandwidth remaining percent) form of LLQ conguration is used, the 75 percent rule no longer applies. This is because the remaining percent
statement refers to the instantaneous residual bandwidth, after the PQ (priority classes) has
been serviced, on the interface. In Example 5-4, up to 33 percent of the link can be dedicated for voice trafc. After voice trafc has been serviced, however, all remaining bandwidth
is divided equally between the MISSION-CRITICAL-DATA class and the default class.
145
Example 5-4 LLQ Policy for VoIP and Multiple Levels of Data Using Remaining Percentage Allocations
policy-map WAN-EDGE
class VOICE
priority percent 33
bandwidth remaining percent 50
class class-default
bandwidth remaining percent 50
LLQ and IPSec

Encrypting a link that carries different classes of trafc and employs LLQ has several
implications:
Encryption happens before queuing. Therefore, classication of all trafc must be

carefully considered, and inner and outer headers must be marked appropriately to
ensure correct classication for egress queuing purposes.
Encryption changes packet sizes because it adds headers to the packet and thus affects
bandwidth allocation for both voice and data trafc. The effect is much more signicant for voice because the packets are relatively small, making the relative overhead
of the new headers far greater. Figure 5-4 illustrates how tunneling and encryption
overhead can more than double the Layer 3 bandwidth required for a voice packet.
The bottleneck within the network node that applies the encryption potentially moves
from the egress interface to the encryption engine. On most routers that have encryption
accelerators, the combined throughput of egress interfaces that can be supported far
exceeds the throughput of the encryption engines. Therefore, encryption for trafc
must be engineered carefully, especially if converged trafc is sent through an
encrypted tunnel.
If encryption is used, especially on the lower-end router platforms, hardware accelerators

should be used. If the egress interface throughput exceeds the throughput of the encryption
engine, packets might heap up waiting to enter the crypto engine; therefore, LLQ must
be applied on the IPSec engine entrance instead of the egress interface, as illustrated in
Figure 5-5. This is a feature named LLQ Before Crypto that became available in Cisco IOS
Software Release 12.2(13)T.
LLQ and cRTP

Traditionally, packet-compression techniques occurred at Layer 2 and, therefore, were not
visible to the LLQ algorithm at Layer 3. LLQ priority class bandwidth thus requires careful
provisioning to reect the actual bandwidth required, after compression but inclusive of
Layer 2 overhead.
146
Figure 5-4
G.729 Voice Bandwidth with GRE and IPSec

IP
Hdr
UDP
RTP
Voice
20
12
20
GRE
IP
Hdr
UDP
RTP
Voice
20
12
20
GRE
IP
Hdr
UDP
RTP
Voice
20
12
20
G.729
60 Bytes
IP GRE
84 Bytes
GRE IP
Hdr
20
IPSec ESP
Tunnel Mode
136 Bytes

Hdr
Hdr
Hdr IV
20
20
ESP
ESP
Pad/NH Auth
2257
12
Encrypted
Authenticated
Figure 5-5
LLQ for IPSec Encryption Engines
VoIP Packets
packets May
mayNeed
need
VoIP
to Be
be Prioritized
prioritized atat
to
cry pto Engine
engine Entrance
entranc e
Crypto
LLQ
Packets
VoIP packets
Prioritized at
prioritized
at Egress
egress
Interface
interface
IPSec
Crypto
Engine
LLQ
Typical Layer 2 data-compression techniques, such as FRF.9 or PPP STAC compression,

tend to have only a mild effect on voice packets. This is because codec compression, such
as G.729, already does very tight compression. Little or no repetitive patterns can be
optimized by applying yet another generic compression technique.
At 40 bytes, the standard IP/RTP/UDP header overhead of a voice packet is quite substantial,
especially when compared to the relatively small size of the payloads (20 bytes for G.729
to 160 bytes for G.711). RTP header compression (or cRTP), which is discussed in greater
147
detail in Chapter 7, Link-Specic Tools, can reduce the overhead requirements of voice
packets from 40 bytes to between 2 and 5 bytes.
A new feature in Cisco IOS Software Release 12.2(13)T allows cRTP to be congured
within the MQC syntax and performed in the CEF switch path. Class-based cRTP also
improves the feedback mechanism to the LLQ policing engine.
LLQ and LFI

Link Fragmentation and Interleaving (LFI) is a QoS technique to minimize serialization
delay on slow (typically less than 768-kbps) WAN links. Essentially, LFI tools chop large
data packets into smaller ones and interleave these data packet fragments with voice
packets. LFI techniques are discussed in more detail in Chapter 7, Link-Specic Tools,
but they have an important relationship with the LLQ mechanism that bears pointing out at
this time.
As shown in Figure 5-2, packets that are assigned to the PQ inside LLQ might escape the
fragmentation engine (the fragmentation engine is represented by the box labeled fragment).
Therefore, trafc with large packet sizes (such as interactive video) should not be assigned
to priority classes on slow-speed links (typically less than 768 kbps). For example, if one
priority class is assigned to VoIP and another is assigned to interactive video on a 512-kbps
link, whenever a 1500-byte video packet is scheduled for transmission, it will require 23 ms
to serialize ((1500 8) / 512,000). This exceeds the per-hop latency target for voice (10 ms)
and nearly consumes the entire one-way jitter allowance target of 30 ms. If a second 1500-byte
IP/VC packet also is scheduled for transmission, voice conversations audibly will become
degraded. Therefore, voice and interactive video never should be assigned dual-LLQ
policies on link speeds less than 768 kbps.
LLQ and ATM PVC Bundles

When different trafc types are converged on a single ATM PVC, the normal LLQ
operation for any interface or PVC applies. However, when PVCs are bundled together,
special considerations must be taken, and these are discussed further in this section.
ATM PVC bundles (previously discussed in Chapter 3, Classication and Marking Tools,
do not require LLQ. Remember that ATM PVC bundles use multiple PVCs to provide
different classes of service over the network backbone. Each PVC is dedicated to a
particular class of trafc. The very nature of LLQ is to separate different classes of trafc
on the same egress interface (or PVC). ATM PVC bundling does just the opposite: It
separates different classes of trafc onto separate, dedicated egress interfaces (or PVCs).
For this reason, LLQ is not supported (or needed) on ATM PVC bundles.
148
For ATM PVC bundles, keep the following points in mind:
LLQ is not supported.
Tx-rings should be set to their default.
LFI is not supported.

cRTP is not supported.
No packet reordering occurs.
All VCs must exist on a single physical interface.
There is no bandwidth sharing. Each PVC is dedicated to trafc of a particular type,
and unused bandwidth cannot be used by trafc of other classes.
LLQ and MLP/Frame Relay Bundles

Other bundling techniques have different QoS characteristics: in particular, MLP and
Frame Relay bundles. These tools aggregate bandwidth across multiple physical interfaces
and bind them together as a larger logical interface. These tools offer additional exibility,
especially with respect to ease of expansion: They simplify management to a large degree.
Interfaces bundled in this manner have the following QoS requirements:
LLQ is supported (and required).
cRTP is supported.
Tx-rings should be set to a minimal value, such as 3.
LFI is supported if the aggregate bandwidth of the bundle is still below 768 kbps. (This
is likely not required because bundling most often allows bandwidths above 768 kbps
so that LFI is no longer necessary.)
Packet reordering occurs.
Multiple physical interfaces are bound together.
There is bandwidth sharing across the aggregate bandwidth of all the links in the
bundle.
Figure 5-6 shows this principle in an MLP bundle aggregating several ATM PVCs spread
across multiple DSL interfaces. The DSL interfaces are terminated by the DSLAM equipment in the network, and the bundled PVCs are delivered on a high-speed ATM interface,
such as an OC-3, at the aggregation site. In this case, LLQ is applied at the MLP bundle
interface, not to the actual physical interfaces.
Figure 5-6
149
MLP PVC Bundling

Remote Site
V
IOS
DSL Drivers
MLP PVC
Bundle
LLQ for Bundle

PQ
ATM
CBWFQ
7200
DSLAM
Aggregation
Site
PVCs
Single or Multiple
DSL Links
LLQ and VoFR

Voice over Frame Relay (VoFR) is an older technique of transporting voice trafc over
Frame Relay networks. It existed before VoIP became implemented widely. VoFR is a pure
Layer 2 method of transporting voice and, for the most part, has been replaced by VoIP in
combination with LLQ.
Cisco IOS Software traditionally has prioritized VoFR packets by default into the same PQ
that is used by LLQ priority classes. The bandwidth allotted to VoFR packets is assigned by
the voice-bandwidth statement in the Frame Relay map class. VoFR packets are funneled
into the PQ of the LLQ algorithm.
NOTE
Care must be taken to never put VoFR and VoIP trafc on the same Frame Relay DLCI on
slow speed links that require fragmentation and interleaving (that is, less than or equal to
768 kbps). This is because VoFR and VoIP require different fragmentation standards
(FRF.11 and FRF.12, respectively). If both VoFR and VoIP are congured on the same
DLCI with fragmentation for each enabled, FRF.11 overrides FRF.12 and disables the
interleaving scheme that VoIP packets require. Obviously then, the quality of VoIP will
deteriorate signicantly.
150
Layer 2 Queuing Tools

Layer 2 queuing tools are required to resolve the prioritization issues generated by multiple
logical interfaces (such as Frame Relay DLCIs and ATM PVCs) as they contend for the
scheduling of a single physical parent interface. These include the following:
Frame Relay Dual-FIFO

PVC Interface Priority Queuing (PIPQ)
Frame Relay Dual-FIFO

On the low-end router non-distributed platforms (Cisco 7200 and lower), Frame Relay
employs a dual-FIFO queuing technique that automatically is invoked at the interface level
when FRF.12, which is discussed in detail in Chapter 7, is congured. FRF.12 depends on
Frame Relay trafc shaping (FRTS) or class-based FRTS being enabled.
In a Frame Relay environment, the Tx-ring does not directly provide back pressure to the
Layer 3 queuing algorithm. Instead, when the Tx-ring is full, it provides back pressure to
the shaper (FRTS or CB-FRTS), which, in turn, signals the Layer 3 queuing system (LLQ)
to engage. Because the FRTS mechanism does not take into account Frame Relay headers
and cyclic redundancy checks (CRCs) in its calculations, it generally is recommended that
you shape to 95 percent of CIR on Frame Relay circuits up to T1/E1 speeds. This, in turn,
engages the LLQ algorithm slightly earlier and improves performance for real-time trafc.
Trafc from each PQ for each DLCI is funneled into the high-priority, dual-FIFO interface
queue; all trafc from the CBWFQ queues from the DLCIs is assigned to the lower-priority,
dual-FIFO interface queue. Thus, the dual-FIFO Layer 2 queues ensure that the priority
class trafc from one DLCI is not delayed by CBWFQ trafc from another DLCI. Frame
Relay Layer 3 and Layer 2 queuing and shaping logic is illustrated in Figure 5-7.
PVC Interface Priority Queuing

Trafc of different importance levels sometimes is separated onto different PVCs for
management or other network design reasons. In such designs, all trafc on the priority
PVC should overtake all trafc on lower-priority PVCs when contending for a single egress
interface. PVC interface priority queuing (PIPQ) is the Layer 2 queuing mechanism that
accomplishes this reordering. It typically is found in service provider scenarios where
dedicated PVCs are used to provide priority services.
PIPQ denes four queues at the interface level. Each PVC on the interface can be assigned
to one of four priorities. As with all PQ techniques, the higher-priority queues have
exhaustive priority over the lower-priority ones and can therefore potentially starve them.
The operation of PIPQ is illustrated in Figure 5-8.
Layer 2 Queuing Tools
Figure 5-7
151
LLQ on Frame Relay PVCs with FRTS and Dual-FIFO

LLQ Policer
VOICE
Call Control
Critical Data
Bulk Data
Best Effort
L3 LLQ/CBWFQ
Classifier
FRTS
L3
Scheduler
Figure 5-8
L2 FR
DualFIFO
Tx-Ring
L1
Scheduler
L2
Scheduler
PVC Interface Priority Queuing (PIPQ) for Frame Relay
Ingress Interface
1
DLCI 101
DLCI 102
DLCI 103
Egress Interface
DLCI 104
DLCI 105
Medium
Normal 4
High
PIPQ
Dequeuing
And
Enqueuing
Low
WAN Circuit
4
In Figure 5-8, DLCIs 101 and 102 are designated as high priority, while DLCIs 103, 104,
and 105 are mapped to the medium, normal, and low queues, respectively. Trafc on DLCIs
101 and 102 will, therefore, have priority over any of the other PVCs, and DLCI 105 will
be capable of sending trafc only if none of the other PVCs has a frame ready to be
transmitted.
152
A practical application for this queuing technique is to place voice trafc on DLCI 101, a
nancial ticker on DLCI 102, and various levels of data trafc on the other three PVCs.
Example 5-5 shows the conguration of PIPQ for DLCI 101.
Example 5-5 Frame Relay PIPQ Example
interface serial0.101
frame-relay interface-queue priority 10 20 30 40
class HIGH-PVC
map-class frame-relay HIGH-PVC
frame-relay interface-queue priority high
Tx-ring
The Tx-ring is a nal FIFO queue that holds frames to be placed immediately on the
physical interface. Its purpose is to ensure that a frame always will be available when the
interface is ready to transmit trafc so that link utilization will be driven to 100 percent of
capacity.
The placement of the Tx-ring is shown on the right of Figures 5-2 and 5-7. The size of the
Tx-ring depends on the hardware, software, Layer 2 media, and queuing algorithm congured on the interface. It is a general best practice to set the Tx-ring to a value of 3 on slowlink interfaces (less than or equal to 768 kbps).
The Tx-ring is especially important on ATM links (including DSL), in which each PVC has
a dedicated driver-level queue (Tx-ring) to ensure adherence to the ATM class-of-service
trafc contract of the PVC. Default ATM Tx-rings are typically deep, containing 64 or
more particles (typically, each particle is 576 bytes) to ensure that enough particles exist in
the buffer to drive the PVC to its full bandwidth utilization. A shallow Tx-ring increases the
number of interrupts and the wait states between the driver and the main CPU. Packets
are downloaded from the Layer 3 queues to the driver for transmission, which is usually
suboptimal at higher speeds.
On the other hand, a deep Tx-ring can impact voice quality. For example, assume that a
50-packet burst of data passes through the LLQ system and into the FIFO Tx-ring. A voice
packet then arrives and is packet number 51 for transmission. The LLQ no longer has the
opportunity to prioritize it over the data packets. A shallow Tx-ring pushes packets back
into LLQ, where prioritization and packet reordering can be accomplished before the
packets are downloaded to the driver. Once in the driver level, the packets have a very
short wait time until transmission.
PAK_priority
153
Thus, the tuning of the Tx-ring is a compromise between optimizing real-time packet
delays and achieving maximum CPU efciency. The higher the bandwidth of the PVC is,
the deeper the Tx-ring can be set without adversely affecting voice quality. For slow links
(less than or equal to 768 kbps), however, a Tx-ring of 3 is recommended to maintain voice
quality.
PAK_priority
Certain trafc types, such as Layer 2 keepalives and Layer 3 routing protocol messages, are
absolutely critical to maintaining network stability. Therefore, it is vital to understand how
the Cisco IOS Software handles these trafc types, especially when provisioning WAN
links that might experience sustained periods of congestion.
By default, Cisco IOS Software (in accordance with RFC 791 and RFC 2474) marks Interior
Gateway Protocol (IGP) trafc (such as Routing Information Protocol [RIP/RIPv2], Open
Shortest Path First [OSPF], and Enhanced Interior Gateway Routing Protocol [EIGRP]) to
IPP6/CS6. These precedence values specify PHBs for these control packets as they pass
through the network. However, Cisco IOS Software also has an internal mechanism for
granting priority to important control datagrams as they are processed through the router.
This mechanism is called PAK_priority.
As datagrams are processed though the router and down to the interfaces, they are encapsulated internally with a small packet header, referred to by an internal label. Within the
elds of this label is a PAK_priority ag that indicates the relative importance of control
packets to the routers internal processing systems. PAK_priority designation is a critical
internal Cisco IOS Software operation and, as such, is not administratively congurable in
any way.
The default internal treatment of PAK_priority datagrams varies slightly between distributed
platforms and nondistributed platforms when MQC policies are bound to the outgoing
interfaces.
Distributed platforms place the PAK_priority datagrams into the class-default queue
and use the PAK_priority ag to avoid dropping the packets.
Nondistributed platforms place PAK_priority datagrams into a separate and dedicated

set of queues (257263) from the class-default, and mark them with a special weight
value (1024).
As noted previously, IGPs are not only marked IPP6/CS6, but they also receive PAK_priority
treatment within the routers. For routing packets that are not marked with IPP6/CS6,
special bandwidth considerations should be given.
154
Additionally, several non-IP control packets, including the following, receive PAK_priority:
Intermediate System-to-Intermediate System (IS-IS) routing protocol messages
High-Level Data Link Control (HDLC) keepalives on serial and POS interfaces
Frame Relay Local Management Interface (LMI) messages
Point-to-Point Protocol (PPP) keepalives on serial and Packet over SONET (POS)
interfaces
ATM operation, administration, and maintenance (OAM) and Address Resolution
Protocol (ARP) cells
Summary
Congestion-management tools were overviewed in this chapter, with an emphasis on the
LLQ algorithm, because this is the principal recommended Cisco IOS queuing tool for
converged networks.
The operation of the LLQ algorithm is important to understand because it directly impacts
QoS design recommendations. Additionally, the interdependence of the QoS toolset is
highlighted because congestion-management tools are shown to be complementary to
classication and marking tools, policing and shaping tools, congestion-avoidance tools,
and link-efciency mechanisms.
Although most queuing policies occur at Layer 3, it is important to consider any impacts
that lower levels of queuing, such as Frame Relay Dual-FIFO, Tx-ring, and PAK_priority,
may have on the overall network design.
Further Reading
Layer 3 Queuing
Conguring priority queuing: http://www.cisco.com/univercd/cc/td/doc/product/

software/ios122/122cgcr/fqos_c/fqcprt2/qcfpq.htm.
Conguring custom queuing: http://www.cisco.com/univercd/cc/td/doc/product/

software/ios122/122cgcr/fqos_c/fqcprt2/qcfcq.htm.
Conguring weighted fair queuing: http://www.cisco.com/univercd/cc/td/doc/

product/software/ios122/122cgcr/fqos_c/fqcprt2/qcfwfq.htm.
IP RTP priority (Cisco IOS Software Release 12.0.5T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios120/120newft/120t/120t5/iprtp.htm.
Further Reading
155
Class-based weighted fair queuing (Cisco IOS Software Release 12.0.5T): http://
cbwfq.htm.
Low-latency queuing (Cisco IOS Software Release 12.0.7T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios120/120newft/120t/120t7/pqcbwfq.htm.
Low-latency queuing for Frame Relay (Cisco IOS Software Release 12.1.2T): http://
dtfrpqfq.htm.
Conguring burst size in low-latency queuing (Cisco IOS Software release 12.1.3T):
121t3/dtcfgbst.htm.
RSVP support for low-latency queuing (Cisco IOS Software Release 12.1.3T): http:/
/www.cisco.com/univercd/cc/td/doc/product/software/ios121/121newft/121t/121t3/
rsvp_llq.htm.
Distributed low-latency queuing (Cisco IOS Software Release 12.1.5T): http://

dtllqvip.htm.
Low-latency queuing with priority percentage support (Cisco IOS Software Release
122newft/122t/122t2/ftllqpct.htm.
Low-latency queuing (LLQ) for IPSec encryption engines (Cisco IOS Software
ios122/122newft/122t/122t13/llqfm.htm.
Layer 2 Queuing
Voice over Frame Relay Queuing Enhancement (Cisco IOS Software Release
120newft/120t/120t5/vofrque.htm.
Frame Relay IP RTP Priority (Cisco IOS Software Release 12.0.7T):

120t/120t7/friprtp.htm.
Frame Relay PVC interface priority queuing (Cisco IOS Software Release 12.1.1T):
121t1/dtfrpipq.htm.
Multilink Frame Relay (FRF.16) (Cisco IOS Software release 12.2.8T):

122t/122t8/ft_mfr.htm.
156
Enhanced Voice and QoS for ADSL and G.SHDSL (Cisco IOS Software Release
122newft/122limit/122y/122yn8/ft_ipqos.htm.
Frame Relay queuing and fragmentation at the Interface (Cisco IOS Software
ios122/122newft/122t/122t13/frfrintq.htm.
Multiclass multilink PPP (Cisco IOS Software Release 12.2.13T):

122t/122t13/ftmmlppp.htm.
Frame Relay PVC interface priority queueing (Cisco IOS Software Release 12.1.1T):
121t1/dtfrpipq.htm.
Understanding and Tuning tx-ring-limit value: http://www.cisco.com/en/US/tech/

tk39/tk824/technologies_tech_note09186a00800fbafc.shtml.
Tx-ring
PAK_priority
How routing updates and Layer 2 control packets are queued on an interface with a
QoS service policy: http://www.cisco.com/warp/public/105/rtgupdates.html.

Explicit congestion notication
CHAPTER
Congestion-avoidance mechanisms, such as Weighted Random Early Detection (WRED),
are complementary to (and dependent on) queuing algorithms. Queuing/scheduling algorithms manage the front of a queue, whereas congestion-avoidance mechanisms manage the
tail of a queue.
Congestion-avoidance QoS tools are designed to handle TCP-based data trafc. TCP has
built-in ow-control mechanisms that operate by increasing the transmission rates of trafc
ows (even though bounded by buffers and window sizes) until packet loss occurs. At
this point, TCP abruptly squelches the transmission rate and gradually begins to ramp the
transmission rates higher again. This behavior makes a strong case against the statement
QoS isnt necessary; just throw more bandwidth at it. If left unchecked, lengthy TCP
sessions (as are typical with bulk data and scavenger applications) can consume any and all
available bandwidth, simply because of the nature of TCP.
When no congestion-avoidance algorithms are enabled on an interface, the interface is said
to tail drop. That is, after the queuing buffers have lled, all other packets are dropped as
they arrive.
In a constricted channel, such as in a WAN or a VPN, all the TCP connections eventually
synchronize with each other as they compete for the channel. Without congestion-avoidance
mechanisms, they all ramp up together, lose packets together, and back off together. This
behavior is referred to as global synchronization. In effect, waves of TCP trafc ow
through the network nodes, with packets overowing the buffers at each wave peak and
lulls in trafc between the waves.
Figure 6-1 illustrates TCP global synchronization behavior attributable to tail-dropping and
the suboptimal effect that this behavior has on bandwidth utilization.
160
Chapter 6: Congestion-Avoidance Tools
Figure 6-1
TCP Global Synchronization

All TCP flows synchronize in waves,
wasting much of the available
bandwidth.
Bandwidth
Utilization
100%
Time
Tail Drop
Three traffic flows start
at different times.
Another traffic flow

starts at this point.
This chapter looks at several variations of the Random Early Detection QoS tool, which
strives to combat this global synchronization behavior exhibited by TCP trafc. Explicit
congestion notication is another tool that can be employed to improve bandwidth utilization
of TCP trafc. The operation and conguration of these tools are discussed in the following
sections.

Random Early Detection (RED) counters the effects of TCP global synchronization by
randomly dropping packets before the queues ll to capacity. Randomly dropping packets
instead of dropping them all at once, as is done in a tail drop, avoids global synchronization
of TCP streams.
Instead of waiting for queuing buffers to ll before dropping packets, RED causes the
router to monitor the buffer depth and perform early discards (drops) on random packets
when the minimum dened queue threshold has been exceeded.
RED drops occur within the operational bounds of TCP retry timers, which slow the
transmission rates of the sessions but prevent them from starting slow. Thus, RED
optimizes network throughput of TCP sessions.
Because UDP does not have any retry logic, congestion-avoidance techniques such as RED
(and variants) do not apply to UDP-based trafc.
161
The Cisco IOS Software does not support RED; it supports only Weighted RED (WRED).
It is important to note that the random-detect Cisco IOS software CLI command enables
WRED. However, if all the packets assigned to an interface or class have the same IPP or
DSCP markings, the effective policy is simply RED.

WRED is an enhancement to RED that enables a degree of inuence over the randomness
of the selection of packets to be dropped. WRED factors the weight, or IPP, of the packet
into the drop-selection process.
Within the WRED algorithm, a minimum threshold for a given IPP value determines the
queue depth at which packets of that IPP value begin to be randomly dropped. The maximum
threshold determines the queue depth at which all packets of that IPP value are dropped.
These thresholds are tunable, as is the mark probability denominator, which determines
how aggressively the packets of a given IPP value are dropped. (For example, a mark
probability denominator value of 10 indicates that up to 1 in 10 packets of a certain
precedence value will be dropped randomlythe maximum rate of 1 in 10 happens at the
maximum threshold, and the drop rate is linear up to this maximum, as shown in Figure 6-2.)
By default, WRED drops packets with lower IPP values sooner than packets with higher
IPP values. A simplied illustration of WRED operation is shown in Figure 6-2.
Figure 6-2
Drop Probability
100%
Queue Depth
Begin
Begin
Begin
Dropping Dropping Dropping
IPP 0
IPP 1
IPP 2
Packets Packets Packets
Drop All
IPP 1
Packets
Drop All
IPP 1
Packets
Drop All
IPP 1
Packets
Max Queue Length
(Tail Drop Everything)
162
WRED can be congured as an interface command, but it is recommended that it be

congured within the MQC syntax because this improves granularity, modularity, and
manageability of the WRED algorithm.
As previously noted, WRED is dependent on queuing. Therefore, before WRED can be
enabled on a class of trafc, a queuing option (usually either bandwidth or fair-queue) must
be enabled on the class. Example 6-1 shows an example of a WRED conguration applied
to the default trafc class.
Example 6-1 WRED Policy Map
class class-default
fair-queue
random-detect enables WRED

DSCP-based WRED was introduced in Cisco IOS Software Release 12.1(5)T. DSCP-based
WRED congures the WRED algorithm to use the AF drop-preference values (as dened
in RFC 2597) of a packets DSCP markings to inuence its drop probability as queues ll.
Remember that the second digit of an AF codepoint indicates drop preference and can range
from 1 (lowest drop preference) to 3 (highest drop preference). For example, if DSCPbased WRED is enabled on an interface, AF23 would be dropped more often (statistically)
than AF22, which, in turn, would be dropped more often (statistically) than AF21.
Figure 6-3 illustrates a simplied (3 PHB) example of DSCP-based WRED.
Figure 6-3
DSCP-Based WRED for AF21, AF22, and AF23
Drop Probability
100%
Queue Depth
Begin
Begin
Begin
Dropping Dropping Dropping
AF23
AF22
AF21
Packets Packets Packets
Drop All
AF23
Packets
Drop All
AF22
Packets
Drop All
AF21
Packets
Max Queue Length
(Tail Drop Everything)
Explicit Congestion Notication
163
The default DSCP-based WRED conguration can be enabled using the dscp-based
keyword of the random-detect command, as shown in Example 6-2.
Example 6-2 DSCP-Based WRED Policy Map
class class-default
fair-queue
random-detect dscp-based enables (RFC 2597) DSCP-based WRED
As with WRED, DSCP-based WRED thresholds and mark probability denominators are
tunable on a per-codepoint basis. In Example 6-3, the minimum threshold is set to begin
dropping AF11-marked packets at 5 (that is, as soon as the queue depth reaches ve packets,
WRED will randomly begin dropping AF11 packets). The maximum threshold for AF11
(at which all AF11-marked packets will be dropped) is set to 20 packets in Example 6-3.
The mark probability denominator is set to 8 (meaning that, between these two thresholds,
up to 1 in 8 packets marked with AF11 will be dropped, and at the maximum threshold of
20, exactly 1 in 8 packets will be droppedabove the maximum threshold of 20, all packets
marked with AF11 will be dropped).
Example 6-3 DSCP-Based WRED Conguration Dropping AF11 packets
Router(config)#policy-map DSCP-WRED
Router(config-pmap)#class class-default
Router(config-pmap-c)#random-detect
random-detect dscp af11 5 20 8

Traditionally, the only way to inform sending hosts that there was congestion on the
network so that the hosts would slow their transmission rates was to drop TCP packets.
RFC 3168, however, dened a new and more efcient way for the network to communicate
congestion to sending hosts: adding explicit congestion notication (ECN) to IP. By
marking the nal 2 bits of the ToS byte of the IP header, devices can communicate to each
other and to endpoints that they are experiencing congestion. These two bits have been
dened as follows:
ECN-Capable Transport (ECT) bitThis bit indicates whether the device supports
ECN.
Congestion Experienced (CE) bitThis bit (in conjunction with the ECT bit)
indicates whether congestion was experienced en route.
Figure 6-4 shows the location of the ECN bits in the TOS byte of an IP packet header.
164
Figure 6-4
IP ToS Byte ECN Bits

ToS
Byte
Version
Length
Len
ID
Offset
TTL
Proto
FCS
IP SA
IP DA
Data
IPv4 Packet
7
ECT CE
ECT Bit:
ECN-Capable Transport
CE Bit:
Congestion Experienced
During periods of congestion, WRED (and DSCP-based WRED) drops packets when the
average queue length exceeds a specic threshold value. ECN is an extension to WRED, in
that ECN marks packets instead of dropping them, to communicate the existence of congestion when the average queue length exceeds a specic threshold value. Routers congured with the WRED ECN feature, introduced in Cisco IOS Software Release 12.2(8)T, use
this marking as a signal that the network is congested. This way, TCP transmission rates
can be controlled without dropping packets, or at least with dropping far fewer packets.
Table 6-1 lists the ECN bit combinations and their meanings.
Table 6-1 ECN Bit Combinations
ECT Bit
CE Bit
Combination Indicates
Not ECN-capable
Endpoints of the transport protocol are ECN-capable
Endpoints of the transport protocol are ECN-capable
Congestion experienced
The bit-setting combinations have the following meanings:
The ECN eld combination 00 indicates that a packet is not using ECN.
The ECN eld combination 11 indicates congestion to the endpoints. Packets arriving
into a full queue of a router are dropped.
The ECN eld combinations 01 and 10, which are called ECT(1) and ECT(0),
respectively, are set by the data sender to indicate that the endpoints of the transport
protocol are ECN-capable. Routers treat these two eld combinations identically.
Data senders can use either or both of these combinations.
165
WRED ECN takes the following actions, depending on the ECN bit settings:
If the number of packets in the queue is below the minimum threshold, packets are
transmitted. This happens whether or not ECN is enabled. This treatment is identical
to the treatment that a packet receives when only WRED is being used on the network.
If the number of packets in the queue is between the minimum threshold and the
maximum threshold, one of the following three scenarios can occur:
If the ECN eld on the packet indicates that the endpoints are ECN-capable
(that is, the ECT bit is set to 1 and the CE bit is set to 0, or the ECT bit is set
to 0 and the CE bit is set to 1) and the WRED algorithm determines that the
packet should be dropped based on the drop probability, the ECT and CE
bits for the packet are changed to 1 and the packet is transmitted. This happens
because ECN is enabled, and the packet gets marked instead of dropped.
If the ECN eld on the packet indicates that neither endpoint is ECN-capable
(that is, the ECT bit is set to 0 and the CE bit is set to 0), the packet can
be dropped based on the WRED drop probability. This is the identical
treatment that a packet receives when WRED is enabled without ECN
congured on the router.
If the ECN eld on the packet indicates that the network is experiencing
congestion (that is, both the ECT bit and the CE bit are set to 1), the packet
is transmitted. No further marking is required.
If the number of packets in the queue is above the maximum threshold, all packets are
dropped and the drop probability is 1. This is the identical treatment that a packet
receives when WRED is enabled without ECN congured on the router, and the
packet ow exceeds its maximum threshold.
WRED ECN is enabled with the ecn keyword on the random-detect command. WRED
ECN cannot be enabled by itself and must be used in conjunction with WRED or DSCPbased WRED (as is shown in Example 6-4).
Example 6-4 WRED-ECN Policy Map Example
Router#sh run
class class-default
fair-queue
random-detect dscp-based
random-detect ecn enables (RFC 3168) ECN-based WRED
166
Summary
Congestion-avoidance tools such as WRED variations and ECN were overviewed in this
chapter.
WRED is used to drop packets randomly (based on congured parameters that control the
drop probabilities) from a TCP session in the network to optimize bandwidth utilization of
TCP trafc. The global synchronization problem of unguarded TCP sessions was
discussed, and the chapter showed how WRED can be used to combat this behavior.
ECN is a newer congestion-avoidance tool, and its operation and conguration was
discussed briey.
Further Reading
DiffServ Standards Relating to WRED
RFC 2597, An Assured Forwarding PHB Group: http://www.ietf.org/rfc/rfc2597.txt.

RFC 3168, The Addition of Explicit Congestion Notication (ECN) to IP: http://
Cisco IOS WRED Documentation
MQC-based WRED (Cisco IOS Software Release 12.0.5T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios120/120newft/120t/120t5/cbwfq.htm.
DiffServ-compliant Weighted Random Early Detection (Cisco IOS Software Release

12.1.5T): http://www.cisco.com/univercd/cc/td/doc/product/software/ios121/121newft/
121t/121t5/dtdswred.htm.
Distributed class-based weighted fair queuing and distributed Weighted Random

Early Detection (Cisco IOS Software Release 12.1.5T): http://www.cisco.com/
univercd/cc/td/doc/product/software/ios121/121newft/121t/121t5/dtcbwred.htm.
WREDexplicit congestion notication (Cisco IOS Software Release 12.2.8T):

122t8/ftwrdecn.htm.
This chapter discusses link-specic QoS tools needed on links of a specic type such as
Frame Relay or MLP, including the following topics:
RTP header compression

Link Fragmentation and Interleaving
CHAPTER
Link-Specic Tools
Link-specic tools refer to mechanisms that are enabled in a point-to-point manner on both
ends of WAN links. These tools combine Layer 2 and Layer 3 features in their functionality.
This chapter covers two types of link-specic tools:
Header-compression techniquesThese mechanisms can reduce bandwidth

requirements for both voice and data trafc.
Link Fragmentation and Interleaving toolsThese mechanisms minimize

serialization delays common to slow-speed WAN links.
The link-specic tools frequently are categorized as QoS tools because they often are used
in conjunction with other QoS techniques, as discussed in this chapter. However, both tools
have application outside the realm of QoS and were developed originally as generic Cisco
IOS Software features, not specically as QoS tools. This chapter limits the discussion to
how these techniques are applied to improve the QoS of the network.
The rst half of this chapter looks at header-compression techniques, which are a suite of
features that reduce the size of the IP packet header. For large data packets, the ratio of the
header bytes to the payload bytes is usually not signicant, and header compression is
seldom a compelling feature. But for short voice packets, the IP (and UDP and RTP)
headers can triple the size of the packet and, therefore, the bandwidth required for a voice
call. These extra bytes also contribute to increased delay, so reducing the size of the header
can provide signicant savings. Although the focus of the following sections is on RTP
header compression, these sections also examine dependencies that this feature has on
related features, such as TCP header compression.
The last half of the chapter explores Link Fragmentation and Interleaving. On links of
768 kbps or higher, this technique has little to no value. However, on slow WAN links,
transmitting a 1500-byte data packet can take long enough that it incurs undue delay in a
voice packet that waits behind it to be transmitted. In these situations, it makes sense to
divide the data packet into smaller segments so that a small voice packet can be interleaved
in between the segments of a large data packet. This bounds the delay incurred on a voice
packet.
170
Chapter 7: Link-Specic Tools
Header-Compression Techniques
One of the key drivers to many VoIP deployments is the nancially attractive bandwidth
savings they deliver. For example, a regular TDM voice call requires a xed allocation
of 64 kbps of bandwidth, whereas a comparable quality VoIP call can be reduced to
approximately 12 kbps. Two types of compression are performed on voice packets to
achieve such bandwidth reductions:
Payload compression, through codecs such as G.729

Header compression, through features such as RTP header compression (also known
as Compressed RTP or cRTP)
G.729 voice compression enables a 64-kbps TDM voice call to be reduced to 8 kbpsa
savings of a factor of 8. The compressed speech samples are carried in VoIP using RTP
packets with 20 bytes of payload per packet (by default). RTP header compression (cRTP)
is a method for making RTP packet headers smaller so that VoIP bandwidth is used more
efciently. cRTP compresses the 40-byte IP/UDP/RTP header of a VoIP packet down to 2
to 5 bytes per packet. This reduces the L3 bandwidth required for a G.729 VoIP call to less
than 12 kbps, as illustrated in Figure 7-1.
Figure 7-1
RTP Header Compression (cRTP)
V=2
CC
PT
Sequence Number
Timestamp
RTP Header:
12 Bytes
Synchronization Source (SSRC) Identifier
Version
Source Port
Destination Port
Length
Checksum
IHL
Type of Service
Identification
Time-To-Live
Total Length
Flags
Protocol
Fragment Offset
Header Checksum
Source Address
Destination Address
Options
UDP Header:
8 Bytes
Padding
cRTP compresses RTP

packet headers from
40 bytes to 2-5 bytes.
IP Header:
20 Bytes
171
Related Standards
Header-compression techniques are described by several standards. The rst headercompression scheme was developed by Van Jacobson to compress IP/TCP headers over
slow-speed serial links. An improvement to this method was offered by the IP header
compression (IPHC) scheme, with an RTP compression extension designed to handle any
IP packet. The most recent addition to the header-compression portfolio is robust header
compression (ROHC), which is a scheme designed for unreliable, high-latency mobile links,
in which bandwidth is very scarce and the links are prone to error (cRTP over satellite
links, a Cisco IOS 12.3[2]T feature, is based on the ROHC compression scheme).
The various IETF RFCs pertaining to header-compression techniques are listed here:
RFC 1144Compresses TCP/IP Headers for Low-Speed Serial Links

RFC 2507IP Header Compression
RFC 2508Compressing IP/UDP/RTP Headers for Low-Speed Serial Links
RFC 2509IP Header Compression over PPP
RFC 3095Robust Header Compression (ROHC)
This is the framework and four proles: RTP, UDP, ESP, and uncompressed. There is also
a Frame Relay standard that describes the use of cRTP on Frame Relay PVCs: FRF.20:
Frame Relay IP Header Compression Implementation Agreement. This standard describes
how IP packets compressed by cRTP should be represented in the Frame Relay frame that
transports them.
TCP Header Compression

As the name implies, TCP header compression (cTCP) is a method of compressing the
headers of TCP/IP segments. However, header compression is much less advantageous, in
terms of the percentage of bandwidth savings, on 1500-byte data packets than cRTP is on
small 20-byte voice packets.
TCP header compression was more relevant in the era when home and remote ofce connections ranged from 9.6 to 19.2 kbps. Although such scenarios are rarely the case today, it
is important to note that the RFCs state that TCP header compression must be enabled automatically whenever cRTP is enabled. This is the case with Ciscos implementations of
cRTP over PPP links.
RTP Header Compression

The capability to support cRTP to gain bandwidth savings for voice streams on slow links
is of paramount importance to private and public VoIP networks. Therefore, cRTP is a key
QoS technique in converged networks with slow-speed edge links. Without cRTP, VoIP
172
service on links slower than about 128 kbps becomes unpractical. Even with cRTP, care
must be taken in the network design that acceptable voice quality can be maintained.
cRTP has the following QoS network design implications:
It changes the packet header and, therefore, must be decompressed before routing can
take place. This makes cRTP a hop-by-hop protocol, with decompression and
recompression occurring in every routing node in which the feature is enabled.
It changes the bandwidth required for a voice call and, therefore, affects provisioning
and engineering of queuing, policing, shaping, and call admission control policies.
It is a compression algorithm and, therefore, CPU intensive. When cRTP is enabled

on a router, care must be taken so that the platform has enough CPU power to perform
this task for the maximum number of calls that might traverse it, and still have enough
CPU for the other data trafc and routing tasks that the CPU must perform. Proof-ofconcept testing for cRTP-enabled congurations is strongly encouraged.
The use of cRTP is dependent on the underlying Layer 2 protocol.
Because cRTP essentially suppresses the Layer 3 and higher headers, it can be used only
on point-to-point Layer 2 connections or links; it cannot be used on broadcast media such
as Ethernet and cable, where reading the packet header is critical for the destination node
to recognize that it is the intended recipient of the frame. Although the RFCs describe the
actual algorithms that header compression protocols use, the protocols are not as much
compression protocols as they are suppression protocols. For example, cRTP works by
keeping the header context (for example, the current values of all the elds in the header)
in both the sender and receiver nodes memory. Compression is based on the principle that
few elds (other than sequence numbers) in the headers change in subsequent packets that
belong to the same ow. cRTP simply omits the transmission of header elds that have not
changed, transmitting instead a reference number or pointer so that the receiving node can
index its memory to nd the actual values of header elds.
During a voice conversation, cRTP does not compress all packets. When the RTP stream
starts up, a few packets have full headers until the compression contexts between the sender
and receiver are established. After that, all packets should be compressed. A full header in
the middle of a conversation is very rare but could be triggered by events such as these:
An event change within the ow cannot be communicated in the compressed header.

One of the samples is too large to be encoded in the compressed header.
An error packet was received, forcing a resynchronization of contexts. It is important to

remember that a few packets at the start of the ow always have full headers, and the
occasional communication of some eld changes also requires full packets with headers.
Nonetheless, the average cRTP stream has a 2- to 5-byte header.
173
Compression Formats
The compression formats that cRTP uses have changed over the years. To achieve backward
compatibility, these are implemented as different options on the interface command, as
follows:
ip rtp header-compression [passive | iphc-format | ietf-format] [periodic-refresh]
The passive option means that the software waits until it receives a compressed packet
before engaging header compression on outgoing RTP packets. This is a legacy command
from the Serial Line IP (SLIP) protocol and is used mainly for dialer situations. PPP
negotiates the use of header-compression regardless of whether the passive option is used.
NOTE
The use of the passive keyword with PPP as the underlying L2 protocol recently has been
disabled to prevent confusion.
The IPHC and IETF formats are discussed in the following sections. The periodic-refresh
option was introduced in Cisco IOS 12.3(2)T Software as part of the robust cRTP over
satellite links feature. It allows a full header to be resent periodically (congurable) to
ensure that compression continues to function over high-loss links.
Cisco Proprietary Format

The original format of cRTP implementation within Cisco IOS was the Cisco proprietary
version that compresses packets only recognized as RTP. RTP packets are recognized by
ensuring that the destination port number is even and within the range of 16385 to 32767
(for audio streams) or 49152 to 65535 (for video streams). The original format of cRTP
allows only up to 256 contexts.
IP Header Compression Format

The IP header compression (IPHC) format was the initial Cisco IOS implementation of
RFC 2507 and RFC 2508. RFC 2507 species the format for TCP and UDP compression,
and RFC 2508 species extensions to RFC 2507specically, to compress RTP packet
headers. In IPHC format over PPP links, TCP and RTP compression are inseparable, as
described by RFC 2509, which species the PPP negotiation for header compression. RFC
2509 states that on PPP, it is impossible to negotiate one type of compression (cTCP or
cRTP) without the other. Thus, when enabling cRTP on a PPP interface, cTCP also automatically is enabled. RFC 2509bis (bis indicates a standards update to the original RFC
2509) decouples cRTP from cTCP but is not yet (fully) implemented within Cisco IOS.
174
However, use of class-based cRTP within the MQC construct can achieve the decoupling
of cTCP and cRTP. This is discussed with additional details later in this chapter. IPHC
format compresses all TCP and UDP packets. Packets that match RTP criteria are compressed using the compressed RTP format, while other packets are compressed using the
compressed non-TCP format. The compressed UDP format is used to compress only RTP
packets that do not t into the compressed RTP template. These formats, and the differences
between them, are discussed in RFCs 2507 and 2508, as follows:
Compressed TCPThe frame contains a datagram with a compressed header with

the format specied in RFC 2507 Section 6a.
Compressed TCP-NODELTAThe frame contains a datagram with a compressed

header with the format specied in RFC 2507 Section 6b.
Compressed non-TCPThe frame contains a datagram with a compressed header

with the format specied in either Section 6c or Section 6d of RFC 2507.
Compressed RTP-8The frame contains a datagram with a compressed header with

the format specied in RFC 2508 Section 3.3.2, using an 8-bit context identier.
Compressed RTP-16The frame contains a datagram with a compressed header

with the format specied in RFC 2508 Section 3.3.2, using a 16-bit context identier.
Compressed UDP-8The frame contains a datagram with a compressed header with

the format specied in RFC 2508 Section 3.3.3, using an 8-bit context identier.
Compressed UDP-16The frame contains a datagram with a compressed header

with the format specied in RFC 2508 Section 3.3.3, using a 16-bit context identier.
IETF Format
The initial Cisco IOS Software implementations of RFCs 2508 and 2509 were not in perfect
compliance with the standards. IETF format, the most recently introduced option (Cisco
IOS 12.3[2]T) for cRTP, attempts to address all of these incompatibilities and to match the
RFCs as closely as possible. In IETF format, as with IPHC format, cTCP and cRTP are
inseparable for PPP links. All TCP and UDP packets are compressed, but only packets recognized as RTP are compressed using the more efcient compressed RTP format. However,
the restriction of recognizing only RTP streams on certain port ranges for compression
has been removed, and cRTP uses any even port over 1024 when the ietf-format keyword
is used.
Layer 2 Encapsulation Protocol Support

The previous sections discussed the standards and formats that govern cRTP in general.
Another dependency of cRTP is on the specic L2 link and the protocol encapsulation
congured for this link.
175
The RFCs tend to discuss only PPP links, yet cRTP is supported over point-to-point links
with HDLC, PPP, Frame Relay, and ATM encapsulations. The implementations over Frame
Relay and HDLC are proprietary, while those over PPP are standards compliant with the
RFCs discussed previously. cRTP on ATM is supported only via PPP or MLP over ATM
(PPPoA or MLPoATM, respectively).
As mentioned earlier, cRTP cannot be supported on broadcast Layer 2 media such as
Ethernet or cable. cRTP works only on point-to-point connections in which there is a single
possible destination for the packet.
The following sections provide additional details about the implementation of cRTP on
each of the link encapsulation types where it is supported.
HDLC
cRTP is implemented over HDLC links in its proprietary version; HDLC itself is a Ciscoproprietary protocol. Cisco format compression is the default, yet IPHC format also is
supported within the CLI. Because the cRTP implementation over HDLC is proprietary,
cTCP and cRTP can be controlled independently (for instance, the implementation does not
have to conform to RFC 2509).On HDLC links, header compression is congured on the
interface using these commands:
ip
ip
ip
ip
tcp
rtp
tcp
rtp
header-compression [passive | iphc-format | ietf-format]

header-compression [passive | iphc-format | ietf-format]
compression-connections [3-256]
compression-connections [3-1000]
The rst two commands turn on cTCP and cRTP, respectively, as discussed earlier. The
compression-connections commands state the maximum number of cTCP or cRTP
sessions that are possible over this link. It is important for this number to match on both
sides of the link, and it should be congured carefully because this governs the amount of
memory set aside to keep the session contexts to compress/decompress the headers. The
defaults and supported ranges have changed over time, so check the Cisco IOS
documentation for the software release that you are running for the correct values.
PPP
The default cRTP format for PPP is IPHC format. PPP negotiates the use of header compression regardless of whether the passive option is used (additionally, the use of passive
with PPP recently has been disabled, to prevent confusion). For IPHC format and IETF
format, turning on either cTCP or cRTP automatically also enables the other type.
On PPP links, header compression is congured at the interface using the same commands
as given previously for HDLC.
176
ATM
cRTP is not supported over native ATM because no standard describes such an implementation. However, a new feature in Cisco IOS 12.2(2)T introduced cRTP over ATM through
PPP over ATM (PPPoATM) and MLP over ATM (MLPoATM). The initial feature release
applied to general ATM interfaces; DSL interface support for this feature followed in
Release 12.3(2)T. cRTP over ATM is congured through PPP using a virtual template. Virtual templates frequently are used with PPP and MLP congurations. They provide a way
to specify a template conguration only once and then apply that template to multiple links,
virtual circuits, or interfaces. This simplies the maintenance of the conguration if there
are many links with similar characteristics.
Any of the congurations given in Example 7-1 can be used for cRTP over ATM links.
Example 7-1 cRTP Conguration for ATM Links
Router#sh run
pvc 0/50
encapsulation aal5mux ppp Virtual-Template1
or
pvc 0/50
encapsulation aal5snap
protocol ppp Virtual-Template1
or
pvc 0/50
encapsulation aal5ciscoppp Virtual-Template1
Coupled with:
!
interface Virtual-Template1
bandwidth 768
ip address 10.200.60.1 255.255.255.252
ip rtp header-compression [passive | iphc-format | ietf-format] [periodic-refresh]
The cRTP conguration (the highlighted text in the example) is the general cRTP command
discussed in an earlier section. Here is it specied within the template Virtual-Template1,
which is, in turn, applied to the ATM PVCs in the example.
Frame Relay
cRTP has been supported on Frame Relay links since the mid-1990s because this
encapsulation is used so frequently on slow-speed access links to remote ofces. The use
of cRTP on Frame Relay has only more recently been standardized as FRF.20 (June 2001).
The implementation of cRTP on Cisco Frame Relay PVCs is currently still proprietary and
will remain so until Cisco implements FRF.20. The current implementation is very close to
the FRF.20 standard but is not exactly the same. cRTP is not supported on IETF PVCs
177
because this support will require FRF.20 compliance. The fact that this is currently a
proprietary implementation also means that cTCP and cRTP can be controlled
independently.
cRTP on PPP over Frame Relay (PPPoFR) and MLP over Frame Relay (MLPoFR) are
standards compliant; this is the same as with the PPP implementations discussed earlier in
this chapter. Therefore, cRTP on PPPoFR is supported on both Cisco and IETF PVCs. If a
standards-compliant implementation of cRTP on Frame Relay is required for a network
design, this form of cRTP should be used.
Although there are many ways to congure cRTP on Frame Relay links, the most common
method is shown in Example 7-2.
Example 7-2 Common cRTP Conguration
Router#sh run
interface Serial 2/0
!
interface Serial 2/0.100 point-to-point
ip address 192.168.0.1 255.255.255.0
frame-relay ip rtp header-compression
The format of the cRTP command is different for Frame Relay than the commands discussed previously. The options on this command include only the passive and periodicrefresh parameters. As discussed earlier, the IETF format of cRTP does not apply to Frame
Relay links.
Frame Relay and ATM Service Interworking

If Frame Relay-to-ATM interworking (FRF.5 or FRF.8) is used in a network and cRTP is
required as well, PPP must be used over the entire link (for example, PPPoATM to PPPoFR,
or MLPoATM to MLPoFR). cRTP is supported over such interworked links as of Cisco IOS
Software Release 12.2(2)T.
Summary of cRTP Formats and Protocol Encapsulations

Table 7-1 summarizes the different formats of cTCP and cRTP that can be used with
different Layer 2 link encapsulations. The table also gives important performance
information in the Switching Path columns. As mentioned earlier, cRTP is CPU intensive;
if it is process-switched, it is signicantly more so. cRTP was implemented in the fast and
CEF switching paths several years ago, but for older releases of Cisco IOS Software, it is
important to check that this is not process-switched before turning on the feature. If it is
process-switched, upgrade to a Cisco IOS release in which cRTP is supported in the fast or
CEF paths.
178
Table 7-1 cRTP Format and Encapsulation Options

Compression Formats
Switching Path
Encapsulation
TCP
Original
RTP
Original
IPHC
Format
IETF
Format
Process
Fast
CEF
dCEF
HDLC
Yes
Yes
Yes
12.3.2T
Yes
Yes
Yes
12.1.5T
PPP
Yes
Yes
12.3.2T
Yes
Yes
Yes
12.1.5T
MLPPP
Yes
Yes
12.3.2T
Yes
Yes
Yes
12.1.5T
Frame Relay
Yes
Yes
Yes
Yes
12.1.5T
PPPoFR
Yes
Yes
12.3.2T
Yes
Yes
Yes
12.2.2T
MLPoFR
Yes
Yes
12.3.2T
Yes
Yes
Yes
12.2.2T
PPPoATM
Yes
Yes
12.3.2T
Yes
Yes
Yes
12.2.2T
MLPoATM
Yes
Yes
12.3.2T
Yes
Yes
Yes
12.2.2T
Yes
Class-Based Header Compression

In Cisco IOS Software Release 12.2(13)T, cRTP was made available within the MQC
syntax, allowing it to be congured on a per-class basis. Class-based cRTP thus improves
the granularity of what trafc is to be compressed, which is a feature often needed to work
around the RFC 2509s requirement that cTCP and cRTP must be negotiated together for
PPP links. Furthermore, class-based cRTP includes cRTP statistics for outbound packets
within the output of the show policy-map interface command, thus improving its manageability. Sample output for this command is shown in Example 7-3. For compression statistics on inbound packets, the show ip rtp header-compression command can be used.
Example 7-3 cRTP Statistics
Router#show policy-map interface Serial 4/1
Serial4/1
Service-policy output:p1
Class-map:class-default (match-any)
30 second offered rate 16000 bps, drop rate 0 bps
Match:any
compress:
header ip rtp
UDP/RTP Compression:
Sent:1000 total, 999 compressed,
41957 bytes saved, 17983 bytes sent
3.33 efficiency improvement factor
99% hit ratio, five minute miss rate 0 misses/sec, 0 max
rate 5000 bps
179
On RFC-compliant software, if cRTP is activated on a certain class of trafc within a service policy, both cTCP and cRTP are applied for the class of trafc (based on the standards
discussed previously in this chapter). However, because TCP and RTP trafc are separated
out into different classes by the classication criteria, usually no TCP trafc is present in
the class where cRTP is enabled. Conversely, there is typically no RTP trafc in a TCP trafc class. Therefore, cRTP effectively can be decoupled from cTCP within the MQC syntax
structure. Such decoupling of the automatic enabling of cTCP whenever cRTP is activated
often is desired for these reasons:
Header-compression ratios on TCP trafc are small, resulting in nominal bandwidth

savings.
Compression algorithms are CPU intensive and should be enabled carefully to

provide the greatest bandwidth returns for the increased CPU cost (for example, on
VoIP, not data).
Data trafc is not real-time sensitive, and header compression does not provide signicant
savings on packets with large payloads. Therefore, cTCP provides negligible packettransmission performance improvements. The match criteria within the class maps determine what trafc is given to the compressor. Decompression, on the other hand, is not controlled by class maps; cRTP decompresses all packets that arrive compressed from the other
side of the link.
The default mode for cRTP inside an MQC class is IPHC format for all encapsulations that
support it, including ATM, PPP, and HDLC. For Frame Relay links, in which IPHC format
is unavailable, the original (Cisco proprietary) format is used as the default. Example 7-4
shows a cRTP conguration inside an MQC class.
Example 7-4 MQC-Based cRTP Conguration
Router#sh run
policy-map CB-CRTP
class VOIP
compression header ip rtp
priority 200
class class-default
fair-queue
In a similar manner, compression header ip tcp can be used for a data trafc class.
The interface conguration for enabling header compression is mutually exclusive with
class-based cRTP. The IPHC format and passive options are autocongured based on the
interface encapsulation and do not appear as being congurable, as shown in Example 7-5.
The number of concurrent connections also is autocongured and calculated based on the
sum of the bandwidths allocated to the classes that have header compression congured
(one connection per 4 kbps of bandwidth available). This must be the same on both ends of
the link.
180
Example 7-5 Class-Based cRTP Conguration

Router(config-pmap-c)#compression header ip ?
rtp configure rtp header compression
tcp configure tcp header compression
Advanced Topics on cRTP

Several other Cisco IOS features should be considered in conjunction with cRTP when
designing converged networks. These are discussed in the following sections.
Tunnels
Although cRTP is an action that works on Layer 3 headers and up, it can be viewed as a
Layer 2 function because it is applied to the packet just before it exits the egress interface.
Therefore, when tunnel congurations (for example, GRE, MPLS, and IPSec) are used,
cRTP ceases to work. (The feature can be congured, but the packets no longer are compressed.) This is because cRTP sees the packet just before it is transmitted on the egress
interface, after the tunnel IP headers have been applied. The cRTP compressor no longer
can recognize the encapsulated packet as an RTP packet and, thus, does not compress it.
RSVP
RSVP is a Layer 3 bandwidth-reservation protocol and generally acts on the Layer 3 packet
size and bandwidth requirements. Therefore, if cRTP is used on a specic voice stream and
RSVP is used to provide the call admission control for that same stream, an over-reservation
of bandwidth will be made for the call because, until recently, RSVP did not take the
bandwidth savings afforded by cRTP into account in its reservations.
As of Cisco IOS Software Release 12.2(15)T, a feedback mechanism between cRTP and
RSVP has been implemented so that RSVP is cognizant of cRTP and adjusts its
reservations appropriately.
RSVP and call admission control technologies are covered in Chapter 8, Bandwidth
Reservation, and Chapter 9, Call Admission Control (CAC).
LLQ, Policing, and Shaping

LLQ, class-based shaping, and class-based (egress) policing all take header compression
into account as of Cisco IOS Software Release 12.2(2)T. This means that class bandwidth
can be congured based on compressed packets. However, keep in mind that decompression occurs on ingress, before any of the input QoS features are activated. Therefore, any
ingress policing policies are applied against uncompressed packets; bandwidth policing
limits should be factored accordingly.
181
Hardware Compression
cRTP is supported only on Cisco PVCs, not on IETF PVCs (unless PPPoFR/MLPoFR is
used). Until Cisco IOS Software Release 12.1(5)T, Frame Relay hardware compression
(FRF.9) was supported on only IETF PVCs, making it difcult to use the optimal compression for all trafc, which is FRF.9 STAC (STAC is the name of the company that rst wrote
the algorithm) for data trafc and cRTP for voice. As of 12.1(5)T, FRF.9 hardware compression also is allowed on Cisco PVCs. Under such a conguration, voice trafc bypasses the
STAC compressor and is compressed by a combination of cRTP and the codec, providing
better compression ratios for voice trafc than STAC compression would.
Performance
Compression techniques, such as cRTP, minimize bandwidth requirements and are highly
useful on slow-speed links. Because of the additional CPU loads these compression
techniques require, they need to be used with caution, especially on WAN aggregation
routers that attach to many remote sites.
In Cisco IOS 12.0, 12.0T, and 12.1 Software, cRTP was process-switched, which rendered
it practically unusable for real networks. In several Cisco IOS releases, between 12.0(7)T
and 12.1(2)T, cRTP was implemented in the fast and Cisco Express Forwarding (CEF)
paths, which dramatically improved its scalability. Nonetheless, cRTP remains one of the
most CPU-intensive QoS features and should be enabled with a careful eye on CPU levels,
especially on large-scale WAN aggregation routers.
Performance is affected positively by the support of class-based cRTP because the trafc
that is compressed now can be ltered more granularly than before. Furthermore, with
class-based cRTP, cTCP use can be avoided, further improving processing efciency.
When the CPU of the node-compressing RTP trafc runs at reasonable levels (for instance,
the target 70 percent), cRTP does not impact delay or voice quality in any adverse manner.

A problem with slow-speed WAN circuits, with respect to voice, is that large data packets
take an excessively long time to be transmitted onto the wire. This delay is referred to as
serialization delay, and it easily can cause a VoIP packet to exceed its delay or jitter tolerance.
Two main tools can be used to mitigate serialization delay on slow links: Multilink PPP
Link-Fragmentation and Interleaving (MLP LFI) and Frame Relay Fragmentation and
Interleaving (FRF.12).
In the context of link-efciency mechanisms, slow-speed links are dened as WAN links
with clock speeds of 768 kbps or less. To offset link-specic delays, larger data packets can
be fragmented and smaller voice packets can be interleaved with the fragments, as shown
in Figure 7-2. This helps reduce the variation in delay caused by a periodic data packet that
182
arrives at the egress interface just before a voice packet does and, therefore, has started
transmission already, causing the voice packets to have to wait until the interface becomes
free before being transmitted itself. Sophisticated queuing techniques such as LLQ do not
help for this problem because this is the result of serialization delay or the elapsed time that
it takes to transmit a packet on the interface. Any subsequent packet must wait for the
interface to become free before starting its own transmission.
Figure 7-2
LFI Operation on Slow-Speed Links
Without LFI
VoIP Packet
1500-Byte Data Packet

214-ms Serialization Delay
for 1500-Byte Frame at 56 kbps
With LFI
Data
Fragment
60 byte
Voice
Data Fragment
VoIP Packet
Data Fragment
LFI tools work at Layer 2 and operate on packets that already have exited the Layer 3 queuing subsystem (LLQ/CBWFQ). As mentioned in Chapter 5, Congestion-Management
Tools, on PPP links, only packets that are not assigned to the PQ of LLQ are fragmented.
This presents an important constraint when provisioning for voice and interactive video on
slow-speed links: Because large video packets assigned to the PQ of the LLQ algorithm are
not fragmented and easily could destroy voice quality, it is not recommended to provision
both voice and interactive video through LLQs on slow-speed links (for instance, those less
than 768 kbps).
Table 7-2 shows the serialization delays of various packet sizes on different link speeds.
Although maximum transmission unit (MTU) size constraints theoretically can be used to
achieve the same purpose as LFI, this is seldom practical because MTU size settings affect
Layer 3 packets and are therefore visible to the end systems. Many end-user server and
desktop applications do not handle MTU size changes elegantly, and even if they do, it is
impractical to change the MTU size on all the servers and user applications in the network.
183
Table 7-2 Serialization Delay by Packet Size and Link Speed

1
Byte
64
Bytes
128
Bytes
256
Bytes
512
Bytes
1024
Bytes
1500
Bytes
56 kbps
143 s
9 ms
18 ms
36 ms
72 ms
144 ms
214 ms
64 kbps
125 s
8 ms
16 ms
32 ms
64 ms
128 ms
187 ms
128 kbps
62.5 s
4 ms
8 ms
16 ms
32 ms
64 ms
93 ms
256 kbps
31 s
2 ms
4 ms
8 ms
16 ms
32 ms
46 ms
512 kbps
15.5 s
1 ms
2 ms
4 ms
8 ms
16 ms
23 ms
768 kbps
10 s
640 s
1.28 ms
2.56 ms
5.1 ms
10.2 ms
15 ms
1536 kbps
5 s
320 s
640 s
1.28 ms
2.56 ms
5.12 ms
7.5 ms
Layer 2 LFI mechanisms are transparent to the Layer 3 IP applications, performing LFI
within the network infrastructure only where necessary. Layer 3 packets are reassembled at
the Layer 2 media edges (requiring LFI), and end systems see normal 1500-byte IP packets.
Fragment Sizes
As a general guideline, a per-hop serialization delay target within the enterprise is smaller
than or equal to 10 ms. Within an SP context, this target might be even tighter.
The following formula can be used to determine the maximum fragment size (in bytes) for
a given jitter target and link speed:
Fragment Size = (Maximum Allowed Jitter in milliseconds [typically 10 ms]) (Link
Speed in kbps) /8
Alternatively, Table 7-3 shows the recommended fragment sizes to achieve a 10-ms
maximum serialization delay for different speed links.
The 768-kbps value to determine when LFI tools are required on a link is derived from the
preceding formula. However, all resulting fragment sizes that are larger than the MTU of
Ethernet (1500 bytes) are shaded in Table 7-3, indicating that no explicit LFI tool is needed
to achieve the 10-ms serialization delay target in the given scenario.
Multilink PPP LFI

Multilink PPP Link Fragmentation and Interleaving (MLP LFI) is a very exible LFI
scheme that can be applied over a single leased line or multiple leased lines, MLP Bundles,
ISDN, Frame Relay circuits (through MLPoFR), ATM PVCs (through MLPoATM), and
ATM-to-Frame Relay Service Interworking (through MLPoFR to MLPoATM SIW).
184
Table 7-3 Maximum Fragment Sizes for 10-ms Serialization Delays

Real-Time Packet Interval
Link or Virtual Circuit Speed
10 ms
56 kbps
70 bytes
64 kbps
80 bytes
128 kbps
160 bytes
256 kbps
320 bytes
512 kbps
640 bytes
768 kbps
960 bytes
1536 kbps
1920 bytes*
Enabling LFI is superfluous whenever the resulting fragment sizes exceed

the MTU of Ethernet (1500 bytes).
When conguring MLP LFI, the only parameter needed is the maximum serialization delay
in milliseconds, which is recommended to be set to 10 ms. The software automatically
calculates the fragment sizes. A sample MLP LFI conguration is shown in Example 7-6.
Example 7-6 MLP LFI Conguration
Router#sh run
interface Multilink1
description 768kbps Leased-Line to RBR-3745-Left
ip address 10.1.112.1 255.255.255.252
ppp multilink
ppp multilink fragment delay 10
ppp multilink interleave
ppp multilink group 1
!
!
interface Serial1/0
bandwidth 786
no ip address
encapsulation ppp
load-interval 30
ppp multilink
!
185
Multiclass Multilink PPP

A limitation of the original implementation of MLP was that only fragments would receive
an MLP header (packets that were smaller than the fragment size would be encapsulated in
a PPP header). Because PPP had no sequencing or reordering capabilities, this presented
reordering limitations when cRTP (which depends on compressed packets arriving in
order) was used in conjunction with MLP bundles over multiple lines, including ISDN.
This issue was xed in Cisco IOS Software Release 12.2(13)T with the introduction of the
Multiclass Multilink PPP feature (MCMP). MCMP encapsulates both fragments and
smaller packets with (sequenced) MLP headers and reassembles all packets in order before
decompression. An example MCMP conguration for an ISDN interface is shown in
Example 7-7.
Example 7-7 MCMP Conguration for an ISDN Interface
Router#sh run
interface BRI0/0
encapsulation ppp
dialer pool-member 1
!
interface Dialer1
encapsulation ppp
dialer pool 1
dialer remote-name routerB-dialer1
dialer-group 1
dialer string 12345678
ppp multilink
ppp multilink fragment-delay 10
ppp multilink links minimum 2
ppp multilink multiclass
The ppp multilink fragment-delay 10 command instructs the Cisco IOS Software to
fragment so that no more than 10-ms delay results. The software calculates the appropriate
fragment size (in bytes) automatically. It is important to add the ppp multilink interleave
statement as well. Although this is a separate command, doing fragmentation without
turning on interleaving has no QoS benet. Both features must be turned on to achieve
the desired operation.
Frame-Relay Fragmentation
Frame-Relay fragmentation, which is dened as FRF.12 for VoIP, operates in a manner very
similar to that of MLP LFI. One important exception, however, is that the conguration of
FRF.12 requires the maximum fragment size to be calculated beforehand and supplied as a
parameter to the command.
186
The following sections discuss several topics that are important to keep in mind when
conguring fragmentation:
Which PVCs to fragment

Which conguration governs that rate that should be used to determine fragment size
The differences between FRF.11 and FRF.12 fragmentation
Which PVCs to Fragment

The purpose of LFI is to break up data frames that share the same transmission interface as
voice packets. If a PVC carries only voice trafc, there is no benet to turning on either
queuing or LFI. However, if other PVCs that share the same physical egress interface carry
data trafc, these PVCs should be fragmented. This is because trafc from all the PVCs
comes together at the physical interface level, and data trafc from one PVC could interfere
with the delay of a voice packet from another PVC. Also, as discussed in Chapter 5,
enabling FRF.12 automatically engages the L2 Dual-FIFO Frame Relay interface queuing
subsystem on the low-end router platforms.
Enabling FRF.12 on a Frame Relay PVC requires the conguration of a map class that is
bound to the DLCI. A sample is shown in Example 7-8.
Example 7-8 FRF.12 LFI Conguration
Router# sh run
interface Serial2/0
no ip address
!
ip address 10.1.121.1 255.255.255.252
description 768kbps FR Circuit to RBR-3745-Left
class FR-MAP-CLASS-768
!
!
map-class frame-relay FR-MAP-CLASS-768
Unlike the fragmentation command for PPP that was discussed in an earlier section, the
frame-relay command expects a byte count as the argument, given as 960 bytes in
Example 7-8. Therefore, it is up to the system administrator to calculate the appropriate
fragment length to achieve a delay that does not exceed 10 ms. Use the information given
in Table 7-3 for this calculation. Also, on Frame Relay links, the frame-relay fragment
187
command automatically enables both fragmentation and interleaving; no separate

command for interleaving is necessary.
In older Cisco IOS releases, fragmentation must be turned on explicitly for each individual
PVC, so it is up to the systems administrator to ensure that all PVCs that share the interface
are fragmented (when necessary, as dictated by the speed of the PVC). In Cisco IOS
Software Release 12.2(13)T, a more convenient feature was introduced that allows CLI at
the Frame Relay interface level to automatically fragment all PVCs on that interface. A
conguration of physical interface-level fragmentation is shown in Example 7-9.
Example 7-9 Interface-Level Fragmentation
Router#sh run
interface serial 0/0
encapsulate frame-relay
NOTE
For ATM, only the PVC that carries combined voice and data trafc on the same VC needs
to be fragmented. ATM inherently interleaves cells from different PVCs, regardless of what
payload they contain.
The Rate That Governs Fragmentation

Although link speed is used as the shorthand term for determining the threshold for LFI,
the shaping rate (CIR) of a Frame Relay PVC, not the physical port speed, really determines
the fragmentation requirement. For example, a 64-kbps PVC on a physical T1 link requires
large data packets be fragmented to 80 bytes to meet the 10-ms delay target. This is because
the shaper, not the clock rate of the physical interface, determines how long a packet will
be held up in the queues before transmission can occur on the egress interface.
FRF.11.1 and FRF.12 Fragmentation

For networks that are migrating from VoFR to VoIP, it must be kept in mind that Frame
Relay LFI actually is covered by two different standards:
FRF.11.1 Annex-C (VoFR)

FRF.12 (VoIP)
A common misconception holds that FRF.12 fragmentation is used to support Voice over
Frame Relay (VoFR), and there is a general unawareness that FRF.11 also species a fragmentation scheme. This causes some misunderstandings about fragmentation for VoFR and
188
VoIPoFR regarding when and how these two voice technologies can be used in the same
network:
FRF.11 and FRF.12 can run on the same PVCThis is not true. PVC runs either
FRF.11 or FRF.12, never both. They are mutually exclusive.
If the PVC is congured for VoFR, it uses FRF.11. If fragmentation is turned
on for this PVC, it uses FRF.11 Annex-C (or the Cisco derivative of this) for
the fragmentation headers.
If the PVC is not congured for VoFR, it uses FRF.3.1 data encapsulation.
If fragmentation is turned on for this PVC, it uses FRF.12 for the
fragmentation headers. Because VoIP is a Layer 3 technology, which is
transparent to Layer 2 Frame Relay, PVCs carrying VoIP use FRF.12
fragmentation.
FRF.12 is used for VoFRThis is partly true. FRF.12 predominantly is used for
VoIP. FRF.12 in a VoFR conguration is used only to fragment data PVCs that share
an interface with a VoFR PVC. FRF.12 never is used on a VoFR PVC itself.
The fragmentation scheme/standard being used does not matter from a CLI point of view.
In all cases, the map class command frame-relay fragment xxx is used. The software
automatically uses the appropriate fragment header type, based on how the DLCI is
congured. However, the fragmentation scheme/standard being used does matter to the
voice network design:
vofr PVCs will not interwork with vofr cisco PVCs.
VoIP and VoFR can be supported on different PVCs on the same interface when
fragmentation is required.
VoIP and VoFR cannot be supported on the same PVC if the speed is such that
fragmentation is required. (VoFR will be ne, but VoIP voice quality will be impaired
because VoIP packets cannot be interleaved between fragments of large data packets.)
FRF.12 (VoIP) fragments voice packets if the fragmentation size is set smaller than the
voice packet size; FRF.11 Annex-C (VoFR) does not fragment voice, no matter what
fragmentation size is congured.
LFI for Frame Relay/ATM Service Interworking

Frame Relay and ATM networks can be interworked in two ways:
FRF.5: network interworking

FRF.8: service interworking
Essentially, FRF.5 is a tunneling protocol, encapsulating an intact Frame Relay frame into
ATM cells and producing an unchanged Frame Relay frame again on the far side of the
network. FRF.8, on the other hand, is a true conversion of the payload of the Frame Relay
frame into the payload of an ATM cell, and it delivers ATM on the far side of the network.
189
Although FRF.5 is not widely used, it can support FRF.12 because the Frame Relay frame
never is changed or disturbed; it merely is encapsulated or tunneled across ATM.
With FRF.8 interworking, a much more common method in backbone networks, FRF.12
cannot be used because there is no means of converting that to an equivalent ATM scheme.
The only way that LFI can be supported end-to-end on Frame Relay/ATM networks using
FRF.8 is to use MLPoFR and MLPoATM and to use MLP LFI over both segments of the
network.
FRF.8 service interworking is a Frame Relay Forum standard for connecting Frame Relay
networks with ATM networks. Service interworking provides a standards-based solution
for service providers, enterprises, and end users. In service interworking translation mode,
Frame Relay PVCs are mapped to ATM PVCs without the need for symmetric topologies;
the paths can terminate on the ATM side. FRF.8 supports two modes of operation of the
interworking function for upper-layer user protocol encapsulation:
Translation modeMaps between ATM and Frame Relay encapsulation. It also

supports the interworking of routed or bridged protocols.
Transparent modeDoes not map encapsulations, but sends them unaltered. This
mode is used when translation is impractical because encapsulation methods do not
conform to the supported standards for service interworking.
MLP for LFI on ATM and Frame Relay service interworking networks is supported for
transparent-mode VCs and translational-mode VCs that support PPP translation (FRF 8.1).
To make MLPoFR and MLPoATM interworking possible, the interworking switch must be
congured in transparent mode, and the end routers must be capable of recognizing both
MLPoFR and MLPoATM headers. This is enabled with the frame-relay interface-dlci
dlci ppp and protocol ppp commands for Frame Relay and ATM, respectively.
When a frame is sent from the Frame Relay side of a connection from ATM to Frame Relay
service interworking, the following should happen to make interworking possible:
The sending router encapsulates a packet in the MLPoFR header.
The receiving router examines the header of the received packet. If the rst 2 bytes of
the received packet are 0x03cf, it treats it as a legal MLPoATM packet and sends it to
the MLP layer for further processing.
In transparent mode, the carrier switch strips off the 2-byte Frame Relay DLCI eld
and sends the rest of the packet to its ATM interface.
When an ATM cell is sent from the ATM side of an ATM to the Frame Relay service
interworking connection, the following should happen to make interworking possible:
The sending router encapsulates a packet in the MLPoATM header.

In transparent mode, the carrier switch prepends the 2-byte Frame Relay DLCI eld
to the received packet and sends the packet to its Frame Relay interface.
190
The receiving router examines the header of the received packet. If the rst 4 bytes
after the 2-byte DLCI eld of the received packet are 0xfefe03cf, it treats it as a legal
MLPoFR packet and sends it to the MLP layer for further processing.
A new standard for ATM to Frame Relay service interworking, FRF.8.1, supports MLP over
ATM and Frame Relay service interworking, but it can be years before all switches are
updated to the new standard.
IPSec Prefragmentation
A feature called prefragmentation for IPSec VPNs was introduced in Cisco IOS Software
Release 12.2(13)T. This often mistakenly is thought to be an LFI feature, but, strictly
speaking, it is not. It does not do LFI, as discussed in this chapter, and its purpose is not to
contain delay of high-priority packets held up behind lower-priority packets on slow-speed
links.
The prefragmentation for IPSec VPNs feature breaks up data packets into two segments.
This is done because a 1500-byte IP packet exceeds the 1500-byte MTU size of the network
when IPSec headers are added onto it, and packets exceeding the MTU size typically are
dropped. This can be xed by setting the application MTU size to 1400 bytes (or less).
However, as explained earlier, changing MTU sizes in a network is cumbersome, at best,
and generally difcult to manage. Therefore, this feature was developed to automatically
segment IPSec frames of more than 1500 bytes into two so that MTU sizes do not need to
be adjusted on the end-host applications to be usable over an IPSec VPN.
As mentioned before, this feature is not to be considered a QoS mechanism for serialization
reduction and tuning purposes. A similar feature for cable/DSL links (typically a voice
VPN), called adjusted maximum segment size, can be turned on with the command ip tcp
adjust-mss. It is recommended that this be congured to 542 on both the outside and inside
interfaces of a cable or DSL device.
Summary
This chapter covered the link-specic tools header compression (cRTP) and Link Fragmentation and Interleaving (LFI). The former typically is used to contain bandwidth use on lowspeed links, while the latter is used to segment large data packets so that a small voice packet
does not have to wait excessively long for the interface to free up.
The different standards governing header compression were discussed, along with the
dependencies among the various mechanisms and the commands to turn on the feature for
different types of link encapsulations.
The last half of the chapter discussed LFI mechanisms for Frame Relay, PPP, and ATM
links. It also covered how to apply LFI for networks in which both Frame Relay and
ATM exist and how translations between the two are done by a carrier network.
Further Reading
191
Further Reading
General
Conguring broadband access: PPP and routed bridge encapsulation conguring PPP
over ATM: http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/
122cgcr/fwan_c/wcfppp.htm.
Enhanced voice and QoS for ADSL and G.SHDSL (12.3.2T): http://www.cisco.com/
univercd/cc/td/doc/product/software/ios123/123newft/123t/123t_2/gtevqos.htm.
IETF Standards
RFC 1141, Incremental Updating of the Internet Checksum:

http://wwwin-eng.cisco.com/RFC/RFC/rfc1141.txt.
RFC 1624, Computation of the Internet Checksum via Incremental Update:

RFC 1144, Compressing TCP/IP Headers for Low-Speed Serial Links:

RFC 2507, IP Header Compression:

RFC 2508, Compressing IP/UDP/RTP Headers for Low-Speed Serial Links:

RFC 1889, RTP: A Transport Protocol for Real-Time Applications:

RFC 2509, IP Header Compression over PPP:

RFC 3095, Robust Header Compression (ROHC): Framework and Four Proles:
RTP, UDP, ESP, and Uncompressed:
Frame Relay Forum Standards
Frame Relay/ATM PVC Network Interworking Implementation Agreement FRF.5:

http://www.frforum.com/5000/Approved/FRF.5/frf.5.pdf.
Frame Relay/ATM PVC Service Interworking Implementation Agreement, FRF.8.1:

http://www.frforum.com/5000/Approved/FRF.8/FRF.8.1.pdf.
Data Compression over Frame Relay Implementation Agreement, FRF.9:

http://www.frforum.com/5000/Approved/FRF.9/frf9.pdf.
192
Voice over Frame Relay Implementation Agreement, FRF.11:

http://www.frforum.com/5000/Approved/FRF.11/frf_11.1.pdf.
Frame Relay Fragmentation Implementation Agreement FRF.12:

http://www.frforum.com/5000/Approved/FRF.12/frf12.pdf.
Frame Relay IP Header Compression Implementation Agreement, FRF.20:

http://www.frforum.com/5000/Approved/FRF.20/FRF.20.pdf.
Header Compression
Express RTP and TCP header compression (Cisco IOS Software Release 12.0.7T):
120t7/rtpfast.htm.
Frame Relay header compression compatibility enhancements (Cisco IOS Software

ios121/121newft/121t/121t2/dtfrhccc.htm.
Distributed Compressed Real-Time Transport Protocol (Cisco IOS Software Release

121newft/121t/121t5/dtdcrtp.htm.
IP RTP coalesce command (Cisco IOS Software Release 12.2.11T):

122t/122t11/ftiprtpc.htm.
Class-based RTP and TCP header compression (Cisco IOS Software Release
122newft/122t/122t13/fthdrcmp.htm.
RSVP support for RTP header compression (Cisco IOS Software Release 12.2.15T):
122t15/ftrsvpcf.htm.
RTP header compression over satellite links (Cisco IOS Software Release 12.3.2T):
123t_2/ftcrtprf.htm.
MLP interleaving and queuing for real-time trafc (Cisco IOS Software Release 12.0):
http://www.cisco.com/univercd/cc/td/doc/product/software/ios120/12cgcr/
dial_c/dcppp.htm#4550.
FRF.12 (Cisco IOS Software Release 12.0.4T): http://www.cisco.com/univercd/cc/td/

doc/product/software/ios120/120newft/120t/120t4/120tvofr/index.htm.
Further Reading
193
FRF.12 support on switched Frame Relay PVCs (Cisco IOS Software Release 12.1.2T):
121t2/dtfragsw.htm.
Link Fragmentation and Interleaving for Frame Relay and ATM virtual circuits (Cisco
IOS Software Release 12.1.5T): http://www.cisco.com/univercd/cc/td/doc/product/
software/ios121/121newft/121t/121t5/dtlfra.htm.
Frame Relay fragmentation with hardware compression (Cisco IOS Software Release
121newft/121t/121t5/dtfrfwhc.htm.
Versatile interface processor-based distributed FRF.12 (Cisco IOS Software Release

121newft/121t/121t5/dtvofrv.htm.
Distributed Link Fragmentation and Interleaving for Frame Relay and ATM interfaces
(Cisco IOS Software Release 12.2.4T): http://www.cisco.com/univercd/cc/td/doc/
product/software/ios122/122newft/122t/122t4/ftdl.htm.
Distributed Link Fragmentation and Interleaving over leased lines (Cisco IOS
Software Release 12.2.8T): http://www.cisco.com/univercd/cc/td/doc/product/
software/ios122/122newft/122t/122t8/ftdl2.htm.
Frame Relay queuing and fragmentation at the interface (Cisco IOS Software Release
122newft/122t/122t13/frfrintq.htm.
Prefragmentation for IPSec VPNs (Cisco IOS Software Release 12.2.13T):

122t/122t13/ftprefrg.htm.
Multiclass Multilink PPP (Cisco IOS Software Release 12.2.13T):

122t/122t13/ftmmlppp.htm.
RSVP as a call admission control mechanism

RSVP and LLQ
RSVP and MPLS trafc engineering
Scalability
RSVP-DiffServ integration
Endpoints and proxies
CHAPTER
The QoS tools discussed so far in this book, including marking, queuing, policing, and
shaping mechanisms, primarily have been DiffServ tools. DiffServ mechanisms provide
bandwidth guarantees (at various level of rigidity), but none of them provides bandwidth
reservations. Guaranteed implies that the bandwidth will be there when needed, but it is
not set aside (or reserved) for a specic application or ow. Reserved, on the other hand,
implies that a ow of packets can be recognized and that a certain amount of bandwidth has
been agreed to be set aside for that ow.
Introduced in Cisco IOS Software Release 11.2, roughly coinciding with the date of many
of the relevant RFCs (summarized for reference at the end of this chapter), RSVP is one of
the oldest Cisco IOS QoS tools. The development and implementation of RSVP precede
the era of converged voice and data networks, yet the purpose of RSVP was always in line
with these later trends: to provide predictable latency and bandwidth guarantees for timesensitive applications. RFC 2205 denes RSVP as follows:
[RSVP is] a resource reservation setup protocol designed for an integrated services
Internet [RFC 1633]. The RSVP protocol is used by a host to request specic qualities
of service from the network for particular application data streams or ows. RSVP is
also used by routers to deliver quality-of-service (QoS) requests to all nodes along the
path(s) of the ows and to establish and maintain state to provide the requested service.
RSVP requests will generally result in resources being reserved in each node along
the data path.
RSVP differs from other QoS tools in the following ways:
It is a signaling protocol.
It reserves resources (bandwidth).
A full implementation requires all nodes in the network to do the following:
Understand the RSVP protocol
Implement a way to reserve resources on that node
196
Chapter 8: Bandwidth Reservation
RSVP Overview
RSVP is a per-ow protocol that requests a bandwidth reservation from every node in the
path of the ow. In its simplest form, RSVP is a unidirectional protocol, so if a bidirectional
reservation is required for a ow, both endpoints must initiate a request for a reservation.
Basic RSVP protocol operation is shown in Figure 8-1 and its conguration in Example 8-1.
The endpoints, or other network devices on behalf of the endpoints, send unicast signaling
messages to establish the reservation: An RSVP PATH message travels outbound requesting the reservation, and an RSVP RESV is returned conrming (or not) that the reservation
was established. The ow can be signaled by an end station (such as Endpoint B in Figure
8-1) or by a router (such as that on behalf of Endpoint A, which is not capable of RSVP
signaling). Every RSVP-enabled router in the path of the ow sees the PATH and RESV
messages and allocates the appropriate queue space for the given ow.
Figure 8-1
Basic RSVP Protocol Operation
Non-RSVP
Capable
Endpoint A
PATH Message
PATH Message
PATH Message
RESV Message
RESV Message
RESV Message
RSVP
Capable
Endpoint B
In summary, the following is true of RSVP:
It is a protocol to signal QoS information to make a reservation of bandwidth.
It maintains state in routers and hosts.
It supports both IPv4 and IPv6.
It can make resource reservations for both unicast and multicast applications.
It is receiver oriented. (In other words, the receiver of a data ow initiates and
maintains the resource reservation used for that ow.)
It is not a routing protocol, but it depends on routing protocols to determine the path
of the ow.
Example 8-1 RSVP Conguration

Router(config)#interface serial 0/1
Router(config-if)#ip rsvp bandwidth 1008 84
RSVP Overview
197
RSVP Service Types

The RSVP protocol denes two distinct service types: controlled load and guaranteed load.
Each is discussed in more detail in the following sections.
Controlled Load
The controlled load service is described by RFC 2211. It provides the ow with soft QoS
(not mathematically bounded, which means that there is no quantitative measure that says
the QoS provided is within, as in a 50-ms delayinstead, it provides an approximate or
qualitative guarantee, which means that the QoS provided is at least as good as the packet
would have gotten otherwise), approximating the service that the same ow would receive
from an unloaded network. The controlled load service uses admission control to ensure
that the service is received even when the network element is overloaded. The controlled
load service includes no quantitative guarantees, and it can be thought of as simple priority
service with admission control. It allows applications to have low delay and high throughput even during times of congestion. For example, adaptive real-time applications, such as
playback of a recorded conference, can use this kind of service.
To ensure that these conditions are met, clients requesting controlled load service provide
the network nodes with an estimation of the data trafc they will generate, which is
described in the trafc specication (or TSpec). The TSpec is one of the parameters in the
RSVP request, as covered in Chapter 1, Introduction to QoS.
Guaranteed Load
The guaranteed load service is described in RFC 2212. It provides the ow with rm
bounds (mathematically provable, which means that explicit, measurable bounds on the
QoS provided to the packet are specied) on end-to-end delays by guaranteeing bandwidth.
Achieving a bounded delay requires that every node in the path supports guaranteed service
or adequately mimics guaranteed service. Guaranteed service allows applications to reserve
bandwidth to meet their requirements. Guaranteed service is invoked by specifying the
trafc (TSpec) and the desired service (RSpec) to the network.
Admission Control
In addition to providing bandwidth reservation to guarantee QoS for a ow, RSVP serves
another purpose of prime importance to real-time ows: call admission control (CAC).
DiffServ tools are highly capable of protecting voice from data (or video from data), but
they fall completely short at protecting voice from voice (or interactive video from
interactive video). For example, if only enough bandwidth is set aside in LLQ for two VoIP
calls, and a third call goes through, the quality of all three calls will deteriorate if only
DiffServ tools are used (without any CAC). However, if a bandwidth reservation is
198
requested before a ow is admitted onto the network, the originating node has the option to
redirect or reject the ow if the reservation fails (because of insufcient bandwidth
availability). To voice and interactive video trafc, the CAC functionality of RSVP is
arguably more critical than the bandwidth reservation functionality (which could be
engineered with DiffServ). CAC through RSVP is discussed in more depth in Chapter 9,
Call Admission Control (CAC).
RSVP and LLQ

RSVP provides admission control. However, to implement the bandwidth and delay
guarantees that RSVP provides for voice trafc, RSVP must work together with LLQ.
The RSVP support for the LLQ feature (introduced in Cisco IOS Software Release
12.1[3]T) allows RSVP to classify voice ows and queue them into the priority queue (PQ)
within the LLQ system while simultaneously providing reservations for non-voice ows
through CBWFQ. The logical ow of this feature is illustrated in Figure 8-2.
Figure 8-2
RSVP and LLQ Logic

Priority
Queue
RSVP
Traffic
Destined
for Interface
Classification
IP RTP
Priority
NonVoice
Flow
Reserved
Queues
Class
Priority
Scheduler
Class 1
Reserved
Queues
Class 2
Reserved
Queues
Class
Default
Reserved
Queues
Default
Unreserved
Queue
Output
MPLS Trafc Engineering
199
In addition to having RSVP enabled on relevant interfaces, this feature requires that trafc
directed to the PQ of LLQ be identied. A built-in voice-like prole can be selected as an
option; in that case, any voice trafc generated by Cisco IOS devices automatically is
assigned to the LLQ. Additionally, this option directs voice trafc from RSVP-enabled
applications, such as Microsoft NetMeeting, to be assigned to the PQ within LLQ. Voicelike trafc means that the trafc ows adhere to certain given arrival rate and packet size
parameters that are derived from real voice ows.
To select the built-in voice-like prole for RSVP to classify trafc into the PQ of LLQ, use
the following command:
Router(config)#ip rsvp pq-profile voice-like

The rst major deployment of RSVP technology came with Multiprotocol Label Switching
(MPLS) trafc engineering (TE). MPLS trafc engineering automatically establishes and
maintains label-switched paths (LSPs) across the backbone by using RSVP to establish and
guarantee tunnels of predictable bandwidth. RSVP operates at each LSP hop and is used
to signal and maintain LSPs based on the MPLS TE calculated path. MPLS TE uses
principles from RFC 2205 (basic RSVP) and RFC 3209 (TE extensions for RSVP) to
accomplish this objective. RSVP used for MPLS TE is illustrated in Figure 8-3. In this
gure, the PATH message requests, for example, 40 Mbps of bandwidth along the path. The
RESV message establishes the bandwidth at each hop and provides the label to use.
Figure 8-3
MPLS TE PATH Setup with RSVP

Router B
Router F
Router E
Router A
Router G
Router C
Router D
= PATH Messages
= RESV Messages
200
Scalability
Discussions about RSVP often are speckled by comments about scalability. Scalability
concerns around RSVP argue that it does not scale well because it keeps per-ow state in
every node, and because the PATH/RESV and refresh messages travel per ow between
every two nodes involved in the path to keep the entire path open and functioning correctly.
For this reason, various scalability enhancements have been introduced into the Cisco IOS
Software, including the following:
Control plane priorityEnsures that the RSVP control messages are not dropped or
unduly delayed, which would cause additional messaging to tear down the path and
re-establish it
Refresh reductionSummarizes the refresh information for all ows on an interface

in a single message instead of sending individual messages per ow
RSVP-DiffServ integration provides a translation between RSVP and DiffServ technologies that is intended to leverage the strengths of each model. RSVP is used for bandwidth
reservation at the edge of the network (where there are fewer ows and the most bandwidth
constraints), but DiffServ is used over the backbone network so that the backbone routers
do not have to keep per-ow states. This topology is shown in Figure 8-4.
Figure 8-4
SP Core Routers Operate in a DiffServ Domain
Enterprise
RSVP (IntServ) Domain
SP DiffServ
Backbone
Enterprise
Enterprise
Enterprise
V
Enterprise
RSVP installed
on interface.
RSVP installed only

to do admission
control.
Further Reading
201
Endpoints and Proxies

Although RSVP as a specication is intended for use along the entire path between the
sender and the receiver, this requires the implementation of RSVP in all endpoints. Most
devices and applications do not comply with this requirement. Additionally, most endpoints
are connected to LANs in which bandwidth is usually plentiful and bandwidth bottlenecks
do not occur; only if the ow traverses the WAN edge router or ISP uplink is congested
typically encountered in the path of the ow.
Summary
This chapter covered RSVP as one of the key IntServ mechanisms in a QoS toolset that is
largely DiffServ oriented. RSVP provides a per-ow specication of QoS through various
parameters and serves as both a call admission control mechanism (allowing or denying a
ow access to the network) and a bandwidth-reservation mechanism. This chapter focused
primarily on the bandwidth reservation aspect of RSVP.
The interactions between RSVP and other QoS tools such as LLQ were discussed, along
with scalability and RSVP-DiffServ integration in networks that are not exclusively IntServ
nor exclusively DiffServ, but that instead use some of the concepts of both worlds. Additionally, for device endpoints that are not capable of RSVP, proxies were discussed to show
how networks with RSVP can be built even for these devices.
Chapter 9 explores RSVPs call admission control capabilities further.
Further Reading
Standards
RFC 2205, Resource Reservation Protocol (RSVP) Version 1 Functional

Specication: http://www.ietf.org/rfc/rfc2205.txt.
RFC 2206, RSVP Management Information Base Using SMIv2:

http://www.ietf.org/rfc/rfc2206.txt.
RFC 2207, RSVP Extensions for IPSec Data Flows: http://www.ietf.org/rfc/

rfc2207.txt.
RFC 2208, Resource ReSerVation Protocol (RSVP) Version 1 Applicability

Statement: Some Guidelines on Deployment: http://www.ietf.org/rfc/rfc2208.txt.
RFC 2209, Resource ReSerVation Protocol (RSVP) Version 1 Message Processing

Rules: http://www.ietf.org/rfc/rfc2209.txt.
202
RFC 2210, The Use of RSVP with IETF Integrated Services: http://www.ietf.org/
rfc/rfc2210.txt.
RFC 2211, Specication of the Controlled-Load Network Element Service:

RFC 2212, Specication of Guaranteed Quality of Service: http://www.ietf.org/rfc/

rfc2212.txt.
RFC 2747, RSVP Cryptographic Authentication: http://www.ietf.org/rfc/

rfc2747.txt.
RFC 2961, RSVP Refresh Overhead Reduction Extensions: http://www.ietf.org/

rfc/rfc2961.txt.
RFC 2998, A Framework for Integrated Services Operation over DiffServ

Networks: http://www.ietf.org/rfc/rfc2998.txt.
RFC 3175, Aggregation of RSVP for IPv4 and IPv6 Reservations: http://
RFC 3209, RSVP-TE: Extensions to RSVP for LSP Tunnels: http://www.ietf.org/

rfc/rfc3209.txt.
General:
http://www.cisco.com/univercd/cc/td/doc/product/software/ios123/123cgcr/qos_r/
qos_i1g.htm#1100504.
RSVP support for low-latency queuing (Cisco IOS Software Release 12.1.3T):
121t3/rsvp_llq.htm.
Multimedia Conference Manager with Voice Gateway Image with RSVP to ATM
SVC Mapping (Cisco IOS Software Release 12.1.5T): http://www.cisco.com/
univercd/cc/td/doc/product/software/ios121/121newft/121t/121t5/dt_mcm5t.htm.
RSVP support for Frame Relay (Cisco IOS Software Release 12.1.5T):
121t5/rsvp_fr.htm.
VoIP call admission control using RSVP (Cisco IOS Software Release 12.1.5T):
121t5/dt4trsvp.htm.
RSVP scalability enhancements (Cisco IOS Software Release 12.2.2T):

122t2/rsvpscal.htm.
Further Reading
203
RSVP support for ATM/PVCs (Cisco IOS Software Release 12.2.2T):

122t2/rsvp_atm.htm.
MPLS DiffServ-aware Trafc Engineering (DS-TE) over ATM (Cisco IOS Software
ios122/122newft/122t/122t8/ft_ds_te.htm.
RSVP refresh reduction and reliable messaging (Cisco IOS Software Release
122newft/122t/122t13/ftrsvpre.htm.
RSVP local policy support (Cisco IOS Software Release 12.2.13T):

122t13/ftrsvplp.htm.
RSVP support for RTP header compression, phase 1 (Cisco IOS Software Release
122newft/122t/122t15/ftrsvpcf.htm.
RSVP message authentication (Cisco IOS Software Release 12.2.15T):

122t/122t15/ftrsvpma.htm.
This chapter discusses the Call Admission Control (CAC) QoS capabilities, and includes
the following topics:
Denition of CAC
Resource Reservations Protocol (RSVP)
Cisco CallManager CAC techniques
H.323 gatekeeper CAC tools
Cisco IOS CAC tools other than RSVP
CHAPTER

Call admission control (CAC) mechanisms are critical to ensuring overall quality for all
trafc types that traverse converged networks, particularly for protecting voice from other
voice trafc, and interactive video from other interactive video trafc.
All the QoS mechanisms discussed so far have dealt with how to treat trafc already
admitted to the network. CAC is the rst toolset that addresses how to keep trafc off the
network (admission) when there are not enough resources to carry the packets. CAC
usually does not apply to data trafc because packets simply can be dropped or delayed
when the network becomes congested or oversubscribed. Real-time trafc can be neither
dropped nor delayed while preserving voice or video quality at the same time. Therefore,
this kind of trafc should be admitted to the network only if it can be carried in a manner
that will provide the required quality.
CAC is an extensive topic in its own right and encroaches upon other areas beyond the
scope of this book, such as call-routing algorithms (dialing plans) and call priority and
preemption capabilities, which typically involve more end-user knowledge than the infrastructure of the network has. A call agent or call server typically makes these types of decisions. On the other hand, the network infrastructure is the only part of the network that has
the knowledge of exact resource availability, such as link status, bandwidth availability, and
trafc congestion points, so the infrastructure is the right place for the basic CAC (admit or
deny) decisions. When a call is not admitted, the call agent or call server can make alternative routing decisionsin other words, the call can be redirected to the PSTN.
The scope of the CAC discussion in this chapter is to provide an overview of the basic CAC
mechanisms and to underscore their importance in QoS design for converged networks.
This chapter examines the three most common methodsRSVP, CallManager locations,
and H.323 gatekeeper CACand provides a brief overview of other Cisco IOSbased
techniques that can be used for aspects of CAC.
CAC Overview
As discussed in Chapter 5, Congestion-Management Tools, and Chapter 6, CongestionAvoidance Tools, data applications are controlled by resolving congestion in the network
instead of applying admission policies. If data applications encounter congestion, packets
206
Chapter 9: Call Admission Control (CAC)
typically are dropped. Protocols such as TCP have inherent detection, recovery, and retransmission logic, which ensures that the session is recovered for the end user. Congestionmanagement policies cannot be applied to real-time applications, such as voice and videoconferencing, and still maintain predictable quality. When the packets that belong to these
applications are admitted onto the network, they cannot be dropped. Traditional TDM
applications such as voice, modem, fax, and video calls assume that bandwidth is available
and do not recover from lost information. Because information arrival is so time sensitive,
there is no point in building error-detection and recovery mechanisms, such as retransmission logic, into these real-time protocols. If the packet cannot be delivered within a small
window of time, it might as well never arrive. Better late than never does not apply to
real-time ows.
CAC Dened
By denition, network bandwidth is nite, and points of congestion do occur. Both realtime and non-real-time trafc types encounter this congestion. If packets cannot be dropped
to resolve congestion, packet ows that cause congestion should not be allowed onto the
network. This makes the case for the deployment of CAC toolsthese are, in essence, the
congestion-avoidance mechanisms for real-time applications. After it is admitted, a realtime ow such as a voice call must be carried; if there arent sufcient bandwidth resources
to carry the ow within the delay and loss bounds of the application, the ow must be
rejected or redirected before it is admitted into the network.
Another way to look at CAC is that most of the QoS tools discussed thus far in this book
strive to protect voice trafc from data trafc. CAC tools protect voice trafc from other
voice trafc (and interactive video from interactive video). For example, if there is sufcient bandwidth provisioned through LLQ to carry only two calls across a link, admitting
a third call will cause packet drops and will impair the voice quality of all three calls in
progress. Such scenarios necessitate the use of CAC to ensure that no more real-time ows
are admitted into the network beyond what the QoS engineering of the nodes allows. This
is illustrated in Figure 9-1.
Formally dened, CAC is a deterministic decision before call establishment on whether the
network resources are available to provide the required QoS to the new call. CAC features
allow VoIP systems to make an informed decision before admitting a new call, based on the
condition of the network. If the call is not admitted, the call can be given the reorder (or
overow) tone or a recorded announcement can inform the caller that the network is too
busy to complete the call attempt. The caller must try again later, or the call can be
redirected to another VoIP route, or the call can be redirected through the PSTN.
CAC Tool Categories
Figure 9-1
207
Protecting Voice from Voice
Example: WAN Bandwidth for 2 Calls Only

Call #1
V
Call #2
VoIP
WAN
Both calls receive good quality because there

is sufficient bandwidth engineered to carry both.
Call #1
Call #2
V
Call #3
VoIP
WAN
When a third call is admitted, all three calls

receive poor quality due to dropped packets
because there is insufficient bandwidth
engineered to carry three calls.
CAC Tool Categories

Three broad categories of CAC tools exist:
LocalThe node makes an admission decision based on local information or

conditions at the node.
Measurement-basedA decision on admitting a new ow is reached based on a

combination of conguration information and reported information of other entities
(such as calls, sessions, ows, bandwidth, memory use, and time slots).
Reservation-basedReservation of resources is performed before admitting the new

ow.
Each of these categories of tools is discussed in the following sections.
208
Local CAC Tools

Local CAC mechanisms perform a voice gateway router function and typically function at
the outgoing gateway. The CAC decision is based on nodal information such as the state of
the outgoing LAN/WAN link that the voice call would traverse if it were allowed to proceed. Clearly, if the local packet network link is down, there is no point in executing complex decision logic based on the state of the rest of the network (because that network is
unreachable). Other local mechanisms include conguration items to disallow more than a
xed number of calls. If the network designer already knows that no more than ve calls
will t across the outgoing WAN links LLQ conguration because of bandwidth limitations, it would be an obvious choice to congure the local gateway node to not allow more
than ve simultaneous calls. The following Cisco IOS features or network design techniques fall in the local CAC mechanisms category:
Physical DS0 limitation

Max-connections
Voice-bandwidth
Trunk conditioning
Local voice busyout (LVBO)
Measurement-Based CAC Tools

Measurement-based CAC techniques look ahead into the packet network to gauge the
state of the network to determine whether to allow a new call. This usually implies sending
probes to the destination IP address (which could be the terminating gateway or endpoint,
or another device in between). The probe returns some measured information on the conditions
that it found while traversing the network to the outgoing (sending) gateway or endpoint.
Typically, loss and delay characteristics are the interesting elements of information for
voice CAC decisions. The outgoing device then uses this information in combination with
congured information to decide whether the network conditions exceed a given or congured threshold. The following Cisco IOS features or network design techniques fall in the
measurement-based CAC mechanisms category, and both use Service Assurance Agent
(SAA) probes:
Advanced voice busyout (AVBO)

PSTN fallback
Resource-Based CAC Tools

Two types of resource-based mechanisms exist: those that calculate needed or available
resources, and those that reserve resources for the call. Resources of interest include link
bandwidth, DSP availability and DS0 time slots on the connecting TDM trunks to a voice
209
gateway, CPU power, and memory. Several of these resources could be constrained at any
nodes or multiple nodes that the call traverses to its destination. The following features or
network design techniques fall in this category:
Gatekeeper resource activity indicator (RAI)

Gatekeeper zone bandwidth
Cisco CallManager Locations CAC
RSVP

Most of the features mentioned in the previous section are Cisco IOS features and pertain
to employing CAC when Cisco IOS voice gateway routers connect traditional telephony
equipment into an IP network. Much of the voice trafc in VoIP networks, however,
originates from IP phones, not from gateways, and none of these features (other than
potentially RSVP) applies to trafc from these devices.
Therefore, Cisco CallManager has additional CAC features to cover the management
of IP phone network deployments. These features are not mutually exclusive. Although
CallManager Locations CAC is deployed in the overall network to manage VoIP bandwidth
availability for both IP phones and voice gateways, a feature such as local or advanced
voice busyout can be deployed at the same time on the voice gateway to push back calls
into the PBX or reroute though the PSTN if IP network conditions do not allow their entry
into the VoIP network.
CallManager Locations CAC is used in a centralized deployment model (centralized CallManager managing phones in both central and remote locations) and is based on the premise
that bandwidth is constrained between locations or sites in the network. Therefore, only a
certain number of calls, regardless of their origination on an IP phone or gateway, can be
allowed into or out of a site at any one time, as shown in Figure 9-2. The premise of this
tool includes an assumption that the bandwidth within a location is unlimited and that constraints exist only between different locations.
Looking at the conguration given in Figure 9-2, Location 2 has WAN bandwidth of
128 kbps available for voice trafc. This does not indicate the total bandwidth into that site,
but just the bandwidth available for voice, as congured into the CallManager at Location 1.
It can represent 50 percent of the total bandwidth into the site and should be correlated to
the amount of bandwidth set aside in the LLQ priority class for Location 2.
210
Figure 9-2
Centralized
CallManager Cluster
Location 2
128 kbps Max
PSTN
V
Router/
GW
IP WAN
Router
Location 1
IP WAN
Location 3
256 kbps Max
CallManager tracks Max Voice

Bandwidth In or Out of each location
and admits or denies a new call
based on this configuration and
current levels of call activity.
IP WAN
Router
The term correlated is used instead of matched because CallManager takes into account
only the uncompressed Layer 3 bandwidth of a call into its call-counting CAC algorithm.
On the other hand, LLQ, as discussed in Chapter 5, also must account for various other
factors, such as cRTP, that affect the bandwidth required for a voice call and must make
provision for the call-signaling bandwidth.
The CallManager Locations CAC feature works in conjunction with CallManager regions
(a feature that controls codec selection between devices that want to communicate), so
bandwidth-constrained links between sites typically are congured to use the G.729 codec
and a counting mechanism within CallManager limits the number of calls into or out of
each site. IP phones and voice gateways are associated with both a region (for codec selection) and a location (for CAC purposes) in the CallManager conguration. For every call
set up, CallManager looks at the source and destination locations of the call and makes a
CAC decision to allow or deny the new call, based on how many calls are already active
between the same two locations and the codecs involved in the calls. Calls conned within
a location are not subject to CAC; as determined by the regions feature, these calls typically
use the G.711 codec.
Gatekeeper CAC
211
Examining the example conguration in Figure 9-2 once more, Location 2 has 128 kbps of
bandwidth available for voice. If the CallManager regions feature is congured so that any
call into or out of Location 2 uses the G.729 codec, 24 kbps of bandwidth are required per
call, and CallManagers locations CAC algorithm will allow a maximum of 5 (128 / 24)
simultaneous calls to Location 2.
CallManager Locations CAC is a heavily deployed feature and is essential for a centralized
deployment in which some IP phones or gateways are separated from each other by WAN
segments of limited bandwidth. CallManager locations should be congured so that calls
between sites do not oversubscribe the bandwidth allocated to the VoIP LLQ in the WAN
routers. CallManager Locations CAC is a simple call-counting mechanism and is unaware
of the topology or state of the network connections. It allows a congured amount of bandwidth (for example, 200 kbps) between two sites. For every call, it subtracts a xed amount
based on the codec selected by the regions feature. Thus, locations CAC works well only
for hub-and-spoke topologies. It is also unaware of L2 protocol overheads (it calculates
purely L3 bandwidth required) and bandwidth-altering features, such as cRTP and tunneling technologies such as GRE and IPSec.
Gatekeeper CAC
Gatekeepers (GK) often are used to arbitrate bandwidth in both CallManager and traditional toll-bypass networks. As with CallManager Locations CAC, gatekeepers employ a callcounting mechanism (based on codec per call) between sites (zones), track bandwidth at an
L3 level, and are also unaware of the network topology. Thus, gatekeeper call admission
control (GK CAC) generally solves the same problem as CallManager Locations CAC and
suffers from the same drawbacks. GK CAC is applicable to a wider set of devices than CallManager Locations CAC, though: GK CAC can be used to arbitrate bandwidth between any
two devices that register with the GK, whereas CallManager Locations CAC can be used
only between IP phones and voice gateways within CallManagers management area. GK
CAC in CallManager networks is used primarily in distributed deployment models to provide CAC on the intercluster trunks or the connections between the CallManager clusters,
as shown in Figure 9-3.
GK CAC also often is deployed in videoconferencing networks to arbitrate bandwidth
among video endpoints, gateways, and MCUs. Unlike CallManager Locations CAC,
the bandwidth subtracted per call is not xed based purely on the codec (CallManager
Locations CAC handles only G.729 and G.711 calls); instead, the bandwidth needed for
the call is requested from the GK as part of the H.225 admissions request traveling from the
endpoint or gateway to the GK.
212
Figure 9-3
Gatekeeper CAC in CallManager Networks

Cisco IOS Gatekeeper
Cisco 3600
WAN
Cisco 3600
Call Placed
V
IP
X1001
Cisco
CallManager 1xxx
PSTN
Voice Overflow
IP
X2002
Cisco
CallManager 2xxx
Call Flow
H.323 RAS
Signaling
RSVP
RSVPs use in providing QoS for real-time ows has a checkered past in the industry, in the
IETF, and in Cisco IOS implementation. However, RSVP is looking increasingly promising
as the only technology that can provide true real-time communications CAC for complex
network topologies and varying trafc patterns. Factors that have contributed to RSVPs
spotty deployment to provide QoS for real-time trafc include the following (most of which
have been resolved):
State-machine synchronization RSVP provides benet to voice calls only if its

state machine and that of the call setup ow are synchronized to provide an RSVP
reservation for the call before the destination partys phone rings. This is referred to
as prering CAC. This synchronization is now in place for H.323 as of Cisco IOS
Software Release 12.1.5T, and for SIP and MGCP as of Cisco IOS Software Release
12.2.8T. The synchronized call setup logic for H.323 is shown in Figure 9-4.
RSVP
Figure 9-4
213
H.323: RSVP Synchronized Call Setup
H.323 Fast Start

Setup includes
UDP port.
H.225 FastStart Set-up

(Outgoing Nodes UDP Ports)
H.225 Call Proceeding
(Terminating Nodes UDP Ports)
RSVP Path
The terminating
node initiates the
first PATH message.
RSVP Resv
RSVP Path
RSVP Resv
RSVP
modification.
RSVP ResvConf
H.225 Alerting or Connect
Originating Node
Alerting/Connect not
sent until an RESV
confirm is received.
Terminating Node
The ow shown in Figure 9-4 depicts only the RSVP messages between the two
communicating endpoints. As discussed in Chapter 8, Bandwidth Reservation, the
RSVP messages actually travel between each two nodes in the network connecting
two endpoints together; each node makes a decision on whether to make or deny the
requested reservation. The ow in Figure 9-4 shows only the end result of the RSVP
messages having traversed all the nodes between the two endpoints.
ScalabilityThese concerns have been addressed with the scalability improvement

and RSVP-DiffServ Integration features discussed in the previous chapter.
Interoperability with other QoS featuresVarious improvements have been made

to RSVP in recent Cisco IOS releases, including control plane prioritization in
Release 12.2.2T and feedback with cRTP in Release 12.2.15T, to ensure that an
accurate amount of bandwidth is reserved and that RSVP works seamlessly with the
other QoS tools.
SecurityTo guard against rogue applications initiating reservations, either for calls
that should not be allowed onto the network or for denial-of-service purposes, RSVP
was enhanced in Cisco IOS Release 12.2.15T. It now does authentication checks on
RSVP messaging to ensure that messages requesting bandwidth are sourced by
legitimate endpoints or nodes and that the requests are for legitimate purposes.
214
Lack of endpoint implementationThe only remaining factor that keeps RSVP

from wide deployment for CAC purposes is its lack of support on IP phones, CallManager, and other voice applications (for example, Unity voicemail) platforms.
However, at the time of writing, development has already begun on RSVP proxy solutions in Cisco IOS that will initiate and broker RSVP reservation across bandwidthconstrained segments of the network on behalf of endpoints such as IP phones. These
Cisco IOS developments also are targeted to address scalability concerns so that
RSVP reservations are made only for segments of the network in which they are
necessary and do not burden the entire network and every call with a reservation.
RSVP addresses many of the shortcomings of the other more narrowly focused CAC
mechanisms:
Only reservations mechanismThe other CAC techniques are call-counting

techniques that cannot guarantee bandwidth. RSVP is the only reservations
mechanism.
Network topology awarenessRSVP is the only CAC technology that can negotiate
a path through any network topology and still guarantee bandwidth on every leg that
the call follows, wherever the routing protocols might point to that path. The other
CAC mechanisms assume simpler hub-and-spoke topologies without alternate
routing capabilities or redundant links or paths they can protect only a virtual leg
between two sites or locations and cannot take into account that some backbone
network segments contain aggregations of calls from several sites. With these
mechanisms, it is still possible to oversubscribe a segment, even if all sites are within
their individual CAC allocations.
Network state awarenessIf multiple links are bound together with technologies
such as MLP bundling, RSVP can react to link failures within the bundle and, therefore, compensate for temporary loss of bandwidth availability in the affected network
segment. The call-counting CAC mechanisms allow bandwidth oversubscription
when failures in the network cause less bandwidth than statically congured to be
available for a period of time.
Accommodation of bandwidth changesIn Cisco IOS Software Release 12.2.2T

and later, RSVP dynamically can adjust the bandwidth allocated to a session or ow
if a higher-layer application requests it.
Free mix of different bandwidth requestsGK CAC and RSVP can do appropriate
CAC for mixes of voice and video calls, each potentially of a varying bandwidth size.
The counting mechanisms, such as CallManager Locations CAC and GK CAC, discussed earlier, fall short of protecting network segments that aggregate calls from
various sites. The complexity of bandwidth arbitration in mixed voice and video networks requires CAC beyond the capabilities of static call countingonly RSVP truly
can protect these environments.
RSVP
215
Example of VoIP CAC Through RSVP

The VoIP call admission control using RSVP feature (Cisco IOS Software Release
12.1[5]T) synchronizes RSVP signaling with H.323 Version 2 signaling to ensure that the
bandwidth reservation is established in both directions before a call moves to the alerting
phase (ringing). This ensures that the called partys phone rings only after the resources for
the call have been reserved. Using RSVP-based admission control, VoIP applications can
reserve network bandwidth and react appropriately if bandwidth reservation fails.
Synchronized RSVP is attempted for an IP call when the requested QoS for the associated
dial peer is set to controlled-load or guaranteed-delay, as long as RSVP has been enabled
for the interface by using the ip rsvp bandwidth command. If the requested QoS level
is set to the default of best effort or RSVP is not enabled, bandwidth reservation is not
attempted. Before Cisco IOS Software Release 12.1(5)T, VoIP gateways used H.323 Version 1 (slow connect) procedures when initiating calls that require bandwidth reservation.
The VoIP CAC through RSVP feature (which is enabled by default) allows gateways to use
H.323 Version 2 (fast connect) for all calls, including those that require RSVP. To enable
backward compatibility, commands are available to force the originating gateway to initiate
calls using slow connect procedures.
If RSVP reservation is attempted but fails, the acceptable QoS for the dial peer determines
the outcome of the call. When the acceptable QoS is congured for best effort, the call setup
proceeds, but without any bandwidth reservation in place. When the acceptable QoS on
either gateway is congured for other than best effort and the RSVP reservation fails, the
call is released. The requested QoS and acceptable QoS are congured through Cisco IOS
Software by using the req-qos and acc-qos dial-peer conguration commands,
respectively.
Table 9-1 summarizes the results of nine call-setup scenarios using fast connect, based on
the QoS levels congured in the VoIP dial peers at the originating and terminating gateways. This table does not include cases in which the requested QoS is best effort and
the acceptable QoS is other than best effort because those congurations are considered
invalid. The following convention is used in the Requested QoS and Acceptable QoS
columns to indicate the conguration for the scenario:
CL: Controlled load

GD: Guaranteed delay
BE: Best effort
The following example, illustrated in Figure 9-5, shows how calls can be made in either
direction between Gateway A and Gateway B, which are connected to POTS phones, with
phone numbers 711 and 712, respectively. The requested QoS indicates that RSVP setup
must complete before the destination phone rings. The acceptable QoS indicates that
the call is released if the RSVP setup fails or does not complete within the allotted time.
Example 9-1 gives the conguration for Gateway A; Example 9-2 gives the conguration
for Gateway B.
216
Table 9-1 Call Results Based on Congured QoS Levels
Figure 9-5
Originating Gateway
Terminating Gateway
Requested
QoS
Acceptable
QoS
Requested
QoS
Acceptable
QoS
CL or GD
CL or GD
CL or GD
CL or GD
Call proceeds only if

both RSVP reservations
succeed.
CL or GD
CL or GD
CL or GD
BE

succeed.
CL or GD
CL or GD
BE
CL or GD
BE
CL or GD
CL or GD

succeed.
CL or GD
BE
CL or GD
BE
Call proceeds regardless

of RSVP results. If
RSVP reservation fails,
call receives best-effort
service.
CL or GD
BE
BE
BE
Call proceeds with besteffort service.
BE
BE
CL or GD
CL or GD
Call is released.
BE
BE
CL or GD
BE
BE
BE
BE
BE
Call is released.
RSVP CAC Synchronization Example

Gateway A
3/0/0
711
Results
Gateway B
10.10.107.107
IP
Network
10.10.107.108
2/0/0
712
RSVP
Example 9-1 Gateway A Conguration

Router# sh run
call rsvp-sync
call rsvp-sync resv-timer 15
!
interface Ethernet0/0
ip address 10.10.107.107 10.255.255.255
service-policy VOIP-LLQ
ip rsvp bandwidth 1000 1000
!
voice-port 3/0/0
!
dial-peer voice 712 voip
destination-pattern 712
session target ipv4:10.10.107.108
req-qos controlled-load
acc-qos controlled-load
!
dial-peer voice 711 pots
port 3/0/0
Example 9-2 Gateway B Conguration

Router# sh run
call rsvp-sync
call rsvp-sync resv-timer 15
!
interface Ethernet0/0
ip address 10.10.107.108 10.255.255.255
service-policy VOIP-LLQ
!
voice-port 2/0/0
!
dial-peer voice 711 voip
session target ipv4:10.10.107.107
req-qos controlled-load
acc-qos controlled-load
!
dial-peer voice 712 pots
port 2/0/0
217
218
Summary
This chapter gave a brief overview of call admission control (CAC) techniques, the QoS
tools designed to keep excess real-time trafc off the network when there isnt bandwidth
to carry it. Packets cannot be dropped summarily (as is done with data trafc) to manage
congestion in the network, so an informed admission decision must be made before
allowing a real-time (typically voice or video) call onto the network.
CAC is a broad subject, and only a high-level overview was discussed here, with particular
emphasis on the most common CAC techniques, including Cisco CallManager Locations
CAC, gatekeeper CAC, and RSVP. RSVP is not deployed widely for CAC in networks
today; however, as support for it increases in more endpoints, this is likely to become the
more prevalent form of CAC over time because it solves many of the shortcomings of the
other techniques discussed in this chapter.
Further Reading
General
Davidson, Jonathan, et al. Deploying Cisco Voice over IP Solutions. Indianapolis:

Cisco Press, 2001.
VoIP call admission control: http://www.cisco.com/en/US/tech/tk652/tk701/

technologies_white_paper09186a00800da467.shtml.
CallManager call admission control: http://www.cisco.com/en/US/products/sw/

voicesw/ps556/products_administration_guide_chapter09186a00800c4cab.html.
RFC 2543, SIP: Session Initiation Protocol: http://www.ietf.org/rfc/rfc2543.txt.
Local voice busyout (Cisco IOS Software Release 12.0.3T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios120/120newft/120t/120t3/busyfm.htm.
Busyout monitor on Cisco 2600 and 3600 series routers (Cisco IOS Software
ios120/120newft/120t/120t7/busy_t7.htm.
Voice busyout enhancements (Cisco IOS Software Release 12.1.2T):

121t/121t2/dt_boenh.htm.
Advanced voice busyout (Cisco IOS Software Release 12.1.3T):

121t/121t3/dt_avbo.htm.
Further Reading
219
PSTN fallback (Cisco IOS Software Release 12.1.3T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios121/121newft/121t/121t3/dtpstnfb.htm.
7200/7500 (Cisco IOS Software Release 12.2.4T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios122/122newft/122t/122t4/ftpstn4t.htm.
RSVP support for low-latency queuing (Cisco IOS Software Release 12.1.3T):
121t/121t3/rsvp_llq.htm.
RSVP support for Frame Relay (Cisco IOS Software Release 12.1.5T):
121t/121t5/rsvp_fr.htm.
VoIP call admission control using RSVP (Cisco IOS Software Release 12.1.5T):
121t5/dt4trsvp.htm.
Control plane DSCP support for RSVP (Cisco IOS Software Release 12.2.2T):
122t2/dscprsvp.htm.
Advanced voice busyout (Cisco IOS Software Release 12.2.4T):

122limit/122x/122xa/122xa_2/ft_cacbo.htm.
Call admission control for H.323 VoIP gateways (Cisco IOS Software Release 12.2.4T):
122limit/122x/122xa/122xa_2/ft_pfavb.htm.
122t8/ft_cac7x.htm.
SIP gateway support of RSVP and TEL URL (Cisco IOS Software Release 12.2.8T):
122limit/122x/122xb/122xb_2/vvfresrv.htm.
MGCP VoIP call admission control (Cisco IOS Software Release 12.2.8T):
122t8/ft_04mac.htm.
122t11/ftcac58.htm.
Enhanced features for local and advanced voice busyout (Cisco IOS Software
ios122/122newft/122t/122t13/ft_lavbo.htm.
220
RSVP support for RTP header compression, phase 1 (Cisco IOS Software
ios122/122newft/122t/122t15/ftrsvpcf.htm.
Measurement-based call admission control for SIP (Cisco IOS Software

ios122/122newft/122t/122t15/ftcacsip.htm.
This chapter discusses the following Catalyst QoS tools:
Classication and marking tools

Mapping tools
Policing and markdown tools
Queuing tools
These tools are considered on a platform-specic basis on the following families of

switches:
Catalyst 2950
Catalyst 2970
Catalyst 3550
Catalyst 3560
Catalyst 3750
Catalyst 4550 (Supervisors II+ through V)
Catalyst 6500 (Supervisor 2 and Supervisor 720)
CHAPTER
10
Catalyst QoS Tools

Most of the QoS tools discussed to this point were covered in the context of Cisco IOS
Software features. As such, these features entail a CPU tax when they are implemented, and
some tools require more CPU cycles than others. The CPU overhead required increases
with the line rates of the media to which the policies are applied.
In a LAN switching environment, QoS policies still are required. (The fallacy of why
simply throwing more bandwidth at it wont solve campus QoS issues is discussed in
more detail in this chapters section discussing campus QoS design.) Implementing such
policies in software at Fast Ethernet, Gigabit Ethernet, and 10-GE speeds would just about
melt any Cisco box. Therefore, Cisco has ported QoS logic from Cisco IOS Software to
hardware ASICs to provision QoS policies at line rates within campus environments.
The porting of QoS from software to hardware has many advantages from a design
perspective, but it also has some caveats (which are discussed later). Some of these design
advantages include the following:
Classication and marking can be ofoaded from routers and moved as close to the
source hosts as administratively possible. This not only extends the DiffServ domain,
but it also enables routers to use their CPU cycles more effectively and efciently.
Trust boundaries can be dened and enforced.
Increased policing options, such as per-port/per-VLAN policing and microow

policing.
Real-time applications, such as voice, can be guaranteed within campus environments

because of preferential/priority hardware queuing.
Congestion can be avoided within the campus, using WRED or similar tools.
Policing can be performed right at the source. This includes both RFC 2597 AF
markdown policing (such as marking down out-of-contract AF11 trafc to AF12,
or even AF13) or simply policing to drop unwanted trafc.
The main caveat that arises from the porting of software QoS to hardware is that QoS
features become hardware specicthat is, they can vary from platform to platform, and
even from line card to line card. This increases the complexity of conguring and managing
campus QoS because many idiosyncrasies need to be kept in mind for each platform or line
card. This caveat further is exacerbated by the fact that some platforms require CatOS,
224
Chapter 10: Catalyst QoS Tools
others require Cisco IOS, and still others (such as the Catalyst 6500) can be congured
using either. To address this issue, common syntaxes, such as the MQC from Cisco IOS,
and command macros, such as AutoQoS, are continuing to be developed to make QoS more
consistent and easier to deploy across Catalyst platforms.
This chapter rst looks at generic QoS models across the Catalyst platform families and
then examines the most currently relevant Catalyst platforms.
Generic Catalyst QoS Models

Platform idiosyncrasies aside for a moment, most Catalyst platforms follow the high-level
generic QoS model shown in Figure 10-1.
Figure 10-1 Generic Catalyst QoS Model
Actions at Ingress
Classification
Generate DSCP
Inspect packet and

determine the DSCP
based on ACLs or
the configuration.
Map the Layer 2
CoS value to a
DSCP value.
Policing
In profile or
out of profile
Compare DSCP to
the configured
policer and
determine if the
packet is in profile or
out of profile.
Actions at Egress
Marking
Based on whether
the packet is in or
out of profile and the
configured parameters.
Determine whether
to pass through,
mark down, or drop
the packet. The
DSCP and CoS are
marked or changed
accordingly.
Queuing and
Scheduling
Based on the CoS,
determine into which
of the egress
queues to place the
packet. Then service
the queues
according to the
configured weights.
The generic Catalyst QoS functions of classication and marking, policing and marking,
and scheduling (congestion management and congestion avoidance) are examined in the
following sections.
Classication, Marking, and Mapping

On most Catalyst platforms (other than the Catalyst 2950), QoS is disabled by default and
must be enabled globally before any function, including classication, occurs.
225
During QoS processing in Catalyst switches, all packets are represented with an Internal
DSCP value. This applies even to non-IP packets because non-IP packets have their
Internal DSCP generated by their CoS values (coupled with the CoS-to-DSCP mapping
table settings) or are assigned the default Internal DSCP value of 0. This Internal DSCP
value is derived from, among other criteria, the trust state of the port. The trust states can
be set to the following:
Trust DSCPThe Internal DSCP is set according to the received DSCP values.
Trust CoSThe Internal DSCP is set based on the received CoS values, through a
default or modied CoS-to-DSCP mapping.
UntrustedThe received CoS is set or reset to the default value (typically 0), and the
Internal DSCP value is set through a default or modied CoS-to-DSCP mapping
(which, likewise, typically results in an Internal DSCP marking of 0).
Trust IP PrecedenceThe Internal DSCP is set based on the received IP Precedence

values, through a default or modied IPP-to-DSCP mapping.
Trust boundaries also can be extended to devices such as IP phones so that the switch will
trust the markings that the IP phone has set. IP phones mark Voice trafc to CoS 5 and
DSCP EF, and Call-Signaling trafc to CoS 3 and DSCP AF31 (or CS3 on newer software
releases).
Modiable CoS-to-DSCP maps provide network administrators granularity in these
mappings. For example, by default, the 3 CoS bits are used as the three most signicant bits
within the DSCP (CoS 5 maps to DSCP 40, which is CS5 and not EF). But when a thirdparty IPT device doesnt mark Voice trafc with CoS and DSCP (as Cisco IP phones do),
the administrator can chose to modify this mapping table to correctly map voice from CoS
5 to DSCP 46 (EF) right at the access switch.
In addition to CoS-to-DSCP maps and IPP-to-DSCP maps, DSCP values themselves can
be mapped to alternate DSCP values, using DSCP mutation maps. Such DSCP-to-DSCP
mutation maps are useful on the borders of two different DiffServ domains.
Classication also can be performed explicitly using access control lists (ACLs), which are
comprised of access control entries (ACEs). Two main ACL types exist: IP and MAC. IP
ACLs usually can use Layer 3 (source or destination addresses) or Layer 4 (TCP/UDP
ports/ranges) for classication purposes. The hardware processes each ACE of an ACL for
a rst-true-match rule. Therefore, it is recommended to order the ACEs to have the most
common matches closer to the top of the ACL to optimize processing.
Figure 10-2 shows a generic Catalyst classication model. Some platforms, such as the
Catalyst 6500, expand on this model, but the main logic remains the same.
226
Figure 10-2 Generic Catalyst Classication, Marking, and Mapping Model

Start
Trust CoS (IP and Non-IP Traffic)
Read ingress interface
configuration for classification. Trust DSCP (IP Traffic)
IP and
Non-IP
Traffic
Use Port
Default
(Non-IP Traffic)
Trust IP
Precedence
(IP Traffic)
Assign DSCP identical
to DSCP in packet.
Check if packet came

with CoS label (tag).
Yes
No
Use CoS
from frame.
(Optional) Modify the

DSCP by using the
DSCP-to-DSCP mutation
map.
Assign default
port CoS.
Generate the DSCP

based on IP precedence
in packet. Use the
IP-Precedence-to-DSCP map.
Generate DSCP from

CoS-to-DSCP map.
Done
Done
Are there any (more) QoS ACLs

configured for this interface?
No
Check if packet came

with CoS label (tag).
Yes
No
Yes
Read next ACL. Is there
a match with a permit action?
No
Yes
Assign the DSCP as
specified by ACL action.
Assign the default

DSCP (0).
Done
Assign the default port

CoS and generate a
DSCP from the
CoS-to-DSCP map.
Generate the DSCP by using

the CoS-to-DSCP map.
Done
227
Policing and Markdown

Three types of policing are supported on Catalyst platforms:
Individual policingApplies bandwidth constraint to each interface

Aggregate policingApplies a bandwidth limit constraint among all interfaces
Microow policing (Catalyst 6500 only)Applies a bandwidth limit constraint to
each ow
As in Cisco IOS, Catalyst policers are based on token-bucket algorithms to determine

in-prole and out-of-prole trafc. Out-of-prole trafc can be either dropped or marked
down. Trafc that is out-of-prole and marked down has both the DSCP and Internal DSCP
marked down to the new value. Figure 10-3 shows a generic Catalyst policing model.
Figure 10-3 Generic Catalyst Policing and Markdown Model
Start
Get the classification

result for the packet.
No
Is a policer configured
for this packet?
Yes
Check if the packet is in
profile by querying the policer.
No
Yes
Pass
Through
Check out-of-profile action

configured for this policer.
Mark
Modify DSCP according to the
policed DSCP map. Generate
a new QoS label.
Done
Drop
Drop packet.
228
Queuing and Dropping

So far, Catalyst QoS might seem remarkably similar to Cisco IOS QoS. However, when it
comes to queuing, the approaches differ to the point that confusion often sets in.
In Catalyst QoS, congestion management (queuing) is performed through a set number of
queues, and congestion avoidance (selective dropping) is performed through a set number
of thresholds per queue. In its most basic form, the number of queues (Q) and thresholds
(T) can be expressed as xQyT. Consider a basic example using two queues with two
thresholds each (2Q2T).
In this Catalyst queuing model, the buffer space for queuing for each line cards port is allocated among two separate queues: Eighty percent of the buffer space (by default) is assigned to
the rst queue, and the remaining 20 percent is assigned to the second queue. This buffer
allocation is depicted in Figure 10-4.
Figure 10-4 Basic Catalyst Queuing: 2Q2T ExampleBuffer Allocation
Queue 2: 20%
Queue 1: 80%
Then two thresholds are dened in each queue. By default, these thresholds are at
80 percent of the queues depth and 100 percent of the queues depth (the tail). These
thresholds are shown in Figure 10-5.
Figure 10-5 Basic Catalyst Queuing: 2Q2T ExampleDefault Thresholds
Queue 2 Threshold 1
Queue 2 Threshold 2
(Queue Tail)
Queue 1 Threshold 2
(Queue Tail)
Front
of
Queues
Queue 1
Threshold 1
On all Catalyst platforms, CoS-to-transmit queue mappings are used to assign packets to
transmit queues (although on some platforms, certain versions of software allow the option
229
to congure DSCP-to-transmit queue mappings). Continuing this basic example, CoS

values 0 through 4 are assigned to the rst queue, while CoS values 5 through 7 are assigned
to the second queue. This is shown in Figure 10-6.
Figure 10-6 Basic Catalyst Queuing: 2Q2T ExampleCoS-to-Queue Mappings
CoS 5-7
CoS 0-4
Additionally, congestion management is overlaid to the queuing model so that each CoS
value (or DSCP value, if the platform/software supports it) can be restricted to a certain
threshold. For example, CoS values 0 and 1 can be restricted to the rst-queue rst-threshold
(1Q1T). If the queue lls past this threshold, all packets with a CoS value of 0 and 1 will
be dropped. Thus, the remainder of the rst queue (1Q2T) is reserved exclusively to hold
only packets with CoS values of 2, 3, and 4. Similarly, CoS values of 6 and 7 are restricted
to the second queues rst threshold (2Q1T), while the remainder of the second queue
(2Q2T) is explicitly reserved for CoS 5.This is shown in Figure 10-7.
Figure 10-7 Basic Catalyst Queuing: 2Q2T ExampleCoS-to-Threshold Mappings
Should the second queue fill beyond this point,
only packets with a CoS value of 5 will be held
(packets with CoS values 6 and 7 will be dropped).
CoS 6+7
2Q1T
CoS 5
CoS 2+3+4
1Q2T
Should the first queue fill beyond this point,

only packets with CoS values 2-4 will be held
2Q2T
CoS 0+1
1Q1T
230
Now that CoS values have been mapped to both queues and thresholds, the scheduling (in
the event of congestion) is performed on this model by a weighted round-robin algorithm.
Such an algorithm favors servicing the second queue (higher CoS values) over servicing the
lower queue. Usually these servicing weights (and the queue sizes) can be overridden from
their default values.
In this manner, preferential service can be offered to voice and network control packets. Yet,
no strict priority servicing is provided to voice, as with Cisco IOS LLQ. An improvement
on the xQyT Catalyst queuing model is the 1PxQyT model, in which a (typically single)
strict-priority queue is added to the model.
As mentioned previously, queues 1 and 2 are serviced by a WRR scheduler, but if any
activity is detected in the priority queue, the WRR scheduler is interrupted, the strictpriority trafc is serviced exhaustively, and the WRR scheduler resumes servicing
queues 1 and 2.
An example 1P2Q2T model, complete with CoS-to-queue and threshold mappings, is
shown in Figure 10-8. In this example, the rst queue is allocated 70 percent of the buffer
space, the second is allocated 15 percent, and the strict priority also is allocated 15 percent,
by default.
Figure 10-8 Basic Catalyst Queuing: 1P2Q2T Example
CoS 5
Should the second queue fill beyond this point,

only packets with CoS values of 6 or 7 will be held
(packets with a CoS value of 4 will be dropped)
1P
CoS 6+7
2Q2T
2
CoS 2+3
1Q2T
Should the first queue fill beyond this point,

only packets with CoS values 2 and 3 will be held
CoS 0+1
1Q1T
CoS 4
2Q1T
0
Catalyst 2950
231
All (new) Catalyst platforms and line cards have some variations of either the xQyT
hardware queuing model or (more likely) the newer 1PxQyT hardware queuing model.
With the generic Catalyst QoS models have been laid out, it is time to examine some
platform-specic applications (and idiosyncrasies) of these models. Only the currentgeneration switches (at the time of writing), including the Catalyst 2950, 3550, 3560, 3750,
4500 (Supervisors II+ through V), and the Catalyst 6500 Supervisor 2 and Supervisor 720
are discussed here.
NOTE
This chapter provides an overview of Catalyst QoS tools only to lay a context for the design
chapters to follow. As such, some platforms have been omitted from the scope of discussion.
For a comprehensive discussion of QoS features on switches not included in this list, such
as the Catalyst 2900XL, 3500XL, 4000-CatOS, and 5000 series switches, refer to Cisco
documentation at http://www.cisco.com/univercd/home/home.htm or to the Cisco Press
book Cisco Catalyst QoS: Quality of Service in Campus Networks, by Mike Flannagan,
Richard Froom, and Kevin Turek.
NOTE
Not every QoS feature that these platforms support is discussed in this chapter; the text
is concerned only with the features that are most relevant to Chapter 12, Campus QoS
Design. All feature discussions in this chapter are based on hardware/software feature sets
available at the time of writing. Refer to Cisco Catalyst documentation for the latest updates
on these platforms and features.
Catalyst 2950
The Catalyst 2950, shown in Figure 10-9, supports only Layer 2 forwarding, making it a
good candidate for a low-end access switch. The main difference between the Catalyst 2950
and the Catalyst 3550 is that the 3550 supports Layer 3 forwarding (IP routing) and thus
can function as a distribution layer switch. Other than that, the platforms are remarkably
similar. Both use the same Cisco IOS code base and have a virtually identical QoS feature
set (with the 3500 boasting a few additional QoS features).
232
Figure 10-9 Cisco Catalyst 2950 Series
Two Cisco IOS software images are supported on the 2950: the Standard Image (SI) and
the Enhanced Image (EI). The full 2950 QoS feature set (including classication, policing,
and marking; mapping, queuing, and scheduling; and AutoQoS) is available as part of the
Enhanced Image. Therefore, it is recommended to use the EI on the 2950 within a QoSenabled campus design. For the remainder of this discussion, it is assumed that the Catalyst
2950 is running an Enhanced Image.
Catalyst 2950 Classication, Marking, and Mapping

Two methods of classication are available on the 2950: on a physical interface basis (no
support exists for VLAN-based classication) and on an MQC/ACL basis.
By default, physical interfaces are untrusted, but they can be congured to trust CoS or
to trust DSCP. At the time of writing, the 2950 does not support the full DSCP range; it
supports only the following values: 0, 8, 10, 16, 18, 24, 26, 32, 34, 40, 46, 48, and 56.
Trust boundaries can be extended to IP phones that are connected to the interface. The
switch uses Cisco Discovery Protocol (CDP) to detect the presence of IP phones.
Catalyst 2950
NOTE
233
In the future, digital certicates will be exchanged between the switch and the IP phone to
verify that an IP phone indeed is connected to the switch and is not a hacker spoong CDP.
In Example 10-1, trust is set for CoS, and the trust boundary has been extended to any
connected IP phones. The range keyword is used to congure multiple interfaces at once.
Example 10-1 Conguring Trust and Trust Extensions on a Catalyst 2950
CAT2950(config)#interface
interface range FastEthernet 0/12 - 24
CAT2950(config-if-range)#mls
mls qos trust cos
mls qos trust device cisco-phone
MQC-based class maps and policy maps also can be used with ACLs to mark trafc on
ingress on the 2950. ACLs supported are standard IP ACLs, extended IP ACLs, and Layer 2
MAC ACLs.
In Example 10-2, an access list identies UDP trafc sourced from an IP/TV server
(10.200.200.200). A class map references this ACL and applies a marking policy to trafc
that matches this criterion, marking these ows to DSCP CS4 (32.)
Example 10-2 Conguring MQC/ACL Classication on the Catalyst 2950
CAT2950(config)#access-list
access-list 100 permit udp 10.200.200.200 255.255.255.255 any
CAT2950(config)#class-map IPTV
CAT2950(config-cmap)#match access-group 100
CAT2950(config-cmap)#exit
CAT2950(config)#policy-map MARK-IPTV-DSCP-CS4
CAT2950(config-pmap)#class IPTV
CAT2950(config-pmap-c)#set
set ip dscp 32
CAT2950(config-pmap-c)#interface FastEthernet 0/1
CAT2950(config-if)#service-policy
service-policy input MARK-IPTV-DSCP-CS4
The default CoS-to-DSCP map for the Catalyst 2950 is shown in Table 10-1. This default
CoS-to-DSCP mapping is actually the same across all Catalyst platforms: The CoS bits are
combined with three trailing zeros (in binary) to form the default DSCP value (in decimal,
the equivalent conversion is achieved by multiplying the CoS value by 8).
Table 10-1
Cisco Catalyst Platforms Default CoS-to-DSCP Map

CoS Value
DSCP Value
16
24
32
40
48
56
234
The default DSCP-to-CoS mapping table for the 2950 is shown in Table 10-2. Notice that
not all DSCP values are included or supported on the Catalyst 2950.
Table 10-2
Catalyst 2950 Default DSCP-to-CoS Map

DSCP Values
8, 10
16, 18
24, 26
32, 34
40, 46
48
56
CoS Values
These mapping values can be modied away from the default settings. As shown in
Example 10-3, CoS 5 is mapped to DSCP 46 (EF), and DSCP CS1 (8) is mapped to CoS 0.
In the rst map (CoS to DSCP), eight separate values must be supplied (in order) that are to
correspond to the CoS values of 0 through 7. Notice that the sixth value (corresponding
to CoS 5) is modied away from the default (40) and explicitly is set to 46 (DSCP EF). With
the second mapping, up to 13 DSCP values (separated by spaces) can be mapped to a CoS
value, which follows the to keyword.
Example 10-3 Conguring Mapping Modications on the Catalyst 2950
CAT2950(config)#mls qos map cos-dscp 0 8 16 24 32 46 48 56
CAT2950(config)#mls qos map dscp-cos 8 to 0
Catalyst 2950 Policing and Markdown

Policing can be applied only on ingress on the Catalyst 2950. The Catalyst 2950 supports
only individual policing. Only 60 policers are supported on Gigabit Ethernet ports, only
6 policers are supported on Fast Ethernet ports on the 2950, and granularity for average
burst is limited to 8-Mbps increments for Gigabit Ethernet ports and 1-Mbps increments for
Fast Ethernet ports.
In Example 10-4, all trafc offered to an interface range (Fast Ethernet 0/2 through 0/4) is
policed to 25 Mbps, beyond which it is dropped.
Example 10-4 Conguring Individual Policing on the Catalyst 2950
CAT2950(config)#access-list
access-list 1 permit any
CAT2950(config)#class-map ANY
CAT2950(config-cmap)#match access-group 1
CAT2950(config)#policy-map POLICE-TO-25MBS
CAT2950(config-pmap)#class ANY
CAT2950(config-pmap-c)#police
police 25000000 16384 exceed-action drop
CAT2950(config)#interface range FastEthernet 0/2 - 4
CAT2950(config-if-range)#service-policy
service-policy input POLICE-TO-25MBS
Catalyst 3550
235
Catalyst 2950 Queuing

The Catalyst 2950 supports four egress queues but no drop thresholds (besides the tail
of the queue). These queues can be congured to schedule using one of two algorithms:
weighted round-robin (WRR) scheduling (4Q1T) or strict-priority scheduling (1P3Q1T).
The default scheduling algorithm is strict priority.
Scheduling weights per queue are set by the wrr-queue bandwidth weight1 weight2
weight3 weight4 command, with weights 1 through 4 representing the weights for each
queue, respectively. The approximate bandwidth allocation per queue is the queues weight
divided by the sum of all weights. When the weight for queue 4 (weight4) is set to 0, the
scheduler is operating in strict-priority fashion, with queue4 enabled as the expedite queue.
In Example 10-5, the scheduler is operating in strict-priority mode, with queue 4 as the
expedite queue (congured by setting weight4 to equal 0). The remaining bandwidth is
divided among queue 1 (25 percent), queue 2 (20 percent), and queue 3 (55 percent). These
percentages are derived from the weight of the queue divided by the sum of all weights
(5 / [5 + 4 + 11]) and (4 / [5 + 4 + 11]) and (11 / [5 + 4 + 11]), respectively.
Example 10-5 Conguring the Expedite Queue and WRR Priority on the Catalyst 2950
CAT2950(config)#wrr-queue bandwidth 5 4 11 0
Finally, CoS values must be assigned to the desired queues. By default, CoS 0 and 1 are
assigned to queue 1, CoS 2 and 3 are assigned to queue 2, CoS 4 and 5 are assigned to queue 3,
and CoS 6 and 7 are assigned to queue 4. Usually, however, it is desirable to send only
CoS 5 (voice) to the expedite queue (queue 4), reassigning CoS 6 and 7 to queue 3. This
modication of the CoS-to-queue mapping can be made with the wrr-queue cos-map
command, as shown in Example 10-6.
Example 10-6 Conguring CoS-to-Queue Mapping on the Catalyst 2950
CAT2950(config)#wrr-queue cos-map 4 5
CAT2950(config)#wrr-queue cos-map 3 6 7
! Assigns CoS 5 to queue 4

! Reassigns CoS 6 and 7 to queue 3
Catalyst 3550
The Catalyst 3550, shown in Figure 10-10, is found in both the access layer and the distribution layer as it supports IP routing. Many of the 3550s QoS features and syntax are identical to the 2950, but a few idiosyncrasies and extra features are unique to the 3550.
236
For example, by default, QoS is disabled on the 3550 and must be enabled manually for any
classication or queuing to occur. Furthermore, enabling QoS on the 3550 on some
versions of software requires disabling IEEE 802.3X ow control on all interfaces, as
shown in Example 10-7.
Example 10-7 Enabling QoS Globally on a Catalyst 3550
CAT3550(config-if-range)#flowcontrol receive off
! Disables flowcontrol
CAT3550(config-if-range)#flowcontrol send off
! Disables flowcontrol
CAT3550(config-if-range)#exit
CAT3550(config)#mls
! Enables QoS globally
mls qos
When ow control has been disabled on all interfaces and QoS has been enabled globally,
all other QoS features become available.
Features and syntax that are identical to the 2950 are not repeated in the following sections;
only features and syntax that are unique to the 3550 are detailed in the following examples.
Catalyst 3550
237

Classication on a per-port level and a per-port, per-VLAN basis is supported on the
Catalyst 3550. Classication also can be performed through port trust states or through
MQC and ACLs. As with the 2950, supported ACL types are IP standard, IP extended, or
a MAC ACL.
Similar to the 2950, the 3550 can trust CoS or DSCP and also limits trust to select devices,
as a Cisco IP phone does. Furthermore, the 3550 supports the capability to trust IP
Precedence, which might be required when interfacing with third-party IP Telephony
devices. Example 10-8 shows how to enable IP Precedence trust on a 3550.
Example 10-8 Conguring IP Precedence Trust on a Catalyst 3550
mls qos trust ip-precedence
Classication through MQC and ACLs is identical to that for the 2950.
The default CoS-to-DSCP map on the 3550 is the same as on all Catalyst platforms
(previously shown in Table 10-1). The default DSCP-to-CoS maps, shown in Table 10-3,
are slightly different on the 3550 than the 2950 because the full DSCP range is supported
on the 3550.
Table 10-3
Catalyst 3500 Default DSCP-to-CoS Map

DSCP Value
0 to 7 8 to 15
16 to 23
24 to 31
32 to 39
40 to 47
48 to 55
56 to 63
CoS Value
Additionally, the 3550 supports an IP Precedence-to-DSCP map, which is essentially

identical to the CoS-to-DSCP map, with the exception that CoS values are replaced with
IP Precedence values.
Mapping modications can be congured identically to those on the 2950.
A new type of mapping, DSCP-to-DSCP mapping, can be congured on a Catalyst 3550.
These maps, known as DSCP mutation maps, join autonomous DiffServ domains. Although
all maps except DSCP mutation maps are dened globally, DSCP mutation maps are
applied on a per-interface basis (specically, on the interface that serves as the border
between DiffServ domains). Actually, more than one DSCP mutation map can be dened
when more than two autonomous DiffServ domains are being joined on a single 3550.
In Example 10-11, PHBs AF21, AF22, and AF23 from one DiffServ domain are mapped to
AF11, AF12, and AF13 in the adjoining DiffServ domain. The interface joining these
DiffServ domains is Gigabit 0/3.
238
Figure 10-11 DSCP Mutation Application Example

QoS Domain 1
QoS Domain 2
Gigabit
Ethernet
0/3
IP Traffic
Catalyst
3550-12T Switch
Gigabit
Ethernet
0/3
Catalyst
3550-12T Switch
Set interface to the DSCP-trusted state.

Configure the DSCP-to-DSCP-mutation map.
The conguration required for such mapping on the border 3550 in DiffServ domain 1 is
Example 10-9 Conguring DSCP Mutation on a Catalyst 3550
CAT3550(config)#mls
mls qos map dscp-mutation DIFFSERV1-TO-DIFFSERV2 18 to 10
CAT3550(config)#mls
CAT3550(config)#mls
CAT3550(config)#int gig 0/3
CAT3550(config-if)#mls qos trust dscp
CAT3550(config-if)#mls
mls qos dscp-mutation DIFFSERV1-TO-DIFFSERV2
A complementary policy would be required on the border switch for the second DiffServ
domain to map in the reverse direction.

The 3550 supports not only ingress policing, but also egress policing on its interfaces.
Furthermore, policers can be dened as individual policers or as aggregate policers.
Aggregate policers are applied not to discrete ows, but cumulatively to all matched trafc
ows. The aggregate rates for such a policer are dened globally.
The 3550 supports 128 policers on Gigabit Ethernet ports and 8 policers on Fast Ethernet
ports. Only one policer can be applied to a packet per direction, and only eight policers can
be dened on an egress port.
Conguring an individual policer is slightly different on a 3550 than on a 2950. This is
because a policed-DSCP map is required to determine the value down to which an out-ofprole DSCP value is to be marked.
Catalyst 3550
239
In Example 10-10, individual Bulk Data trafc ows (marked on a data center server
connected to Fast Ethernet 0/12) in excess of 10 Mbps are marked down from AF11
(DSCP 10) to AF12 (DSCP 12).
Example 10-10 Conguring an Individual Policer on a Catalyst 3550
CAT3550(config)#mls
mls qos map policed-dscp 10 to 12
CAT3550(config)#class-map match-any BULK
CAT3550(config-cmap)#match ip dscp af11
CAT3550(config)#policy-map BULK-MARKDOWN
CAT3550(config-pmap)#class BULK
police 10000000 8000 exceed-action policed-dscp-transmit
CAT3550(config-pmap-c)#interface fa 0/12
CAT3550(config-if)#service-policy input BULK-MARKDOWN
The Catalyst 3550 is the only switch (at the time of writing) to support per-port/per-VLAN
policing (in the ingress direction). Per-port/per-VLAN policing requires an additional class
map to be dened with a match-all clause and two match statements: one for the VLAN
to be matched and the other for the class map to apply. For example, if the preceding Bulk
Data policing policy is to be applied only on VLAN 10, the per-port/per-VLAN policy
would look like Example 10-11.
Example 10-11 Conguring a Per-Port, Per-VLAN Individual Policer on a Catalyst 3550
CAT3550(config)#mls qos map policed-dscp 10 to 12
CAT3550(config-cmap)#match ip dscp af11
CAT3550(config-cmap)#class-map
class-map match-all VLAN10-BULK
CAT3550(config-cmap)#match
match vlan 10
CAT3550(config-cmap)#match
match class-map BULK
CAT3550(config-cmap)#policy-map VLAN10-BULK-MARKDOWN
CAT3550(config-pmap)#class VLAN10-BULK
CAT3550(config-pmap-c)#police 10000000 8000 exceed-action policed-dscp-transmit
CAT3550(config-pmap-c)#interface fa 0/12
CAT3550(config-if)#service-policy input VLAN10-BULK-MARKDOWN
In addition to individual policers, an aggregate policer can be applied to an access-todistribution uplink on egress to mark down cumulatively out-of-prole trafc. Building on
the previous example, all Bulk Data trafc in excess of 100 Mbps (regardless of how many
ows are involved) is marked down to AF12 on the Gigabit 0/2 uplink. The policed-DSCP
map does not need to be redened, but it can be leveraged by both individual and aggregate
policers, as shown in Example 10-12.
Example 10-12 Conguring an Aggregate Policer on a Catalyst 3550
mls qos aggregate-policer BULK-MARKDOWN-AGG 1000000000 8000
exceed-action policed-dscp-transmit
CAT3550(config)#policy-map BULK-MARKDOWN-AGG
continues
240
Example 10-12 Conguring an Aggregate Policer on a Catalyst 3550 (Continued)

police aggregate BULK-MARKDOWN-AGG
CAT3550(config-pmap-c)#int gig 0/2
CAT3550(config-if)#service-policy
service-policy output BULK-MARKDOWN-AGG
Catalyst 3550 Queuing and Dropping

The 3550 has four egress queues, one of which (queue 4) can be congured as a strictpriority queue that (unlike the 2950) is not enabled by default but that needs to be enabled
on a per-interface basis with the priority-queue out command. An example of enabling the
strict-priority queue on a Gigabit Ethernet uplink is shown in Example 10-13.
Example 10-13 Conguring a Strict-Priority Queue on a Catalyst 3550
CAT3550(config-if)#priority-queue
priority-queue out
As with the 2950, weights can be congured to allocate bandwidth to each queue by the
weighted round-robin scheduler. The syntax is identical to that of the 2950, except that if a
strict-priority queue has been enabled on an interface, the value for weight4 simply is
ignored.
NOTE
Queue limits also can be tuned away from defaults using the wrr-queue queue-limit
command. The default queue limits are such that each queue is assigned 25 percent of the
available buffer space.
It should be noted that when the queue limits are modied, the queue temporarily is shut
down during the hardware reconguration, and the switch may drop newly arrived packets
destined to the queue.
CoS values are assigned to their respective queues using the wrr-queue cos-map
command. A difference between the Catalyst 3550 and the 2950 is that the CoS-to-queue
maps are dened on a per-interface basis on the 3550 (not globally, as on a Catalyst 2950).
By default, CoS 0 and 1 are assigned to queue 1, CoS 2 and 3 are assigned to queue 2, CoS
4 and 5 are assigned to queue 3, and CoS 6 and 7 are assigned to queue 4. Usually, it is
desirable to send only CoS 5 (voice) to the expedite queue (queue 4) and to reassign CoS 6
and 7 to queue 3. This modication of the CoS-to-queue mapping can be made with the
wrr-queue cos-map interface command on the 3550, as shown in Example 10-14.
A QoS feature that the 3550 enjoys that the 2950 does not is the capability to support
congurable drop thresholds per queue on Gigabit Ethernet ports. Therefore, Catalyst 3550
Catalyst 3550
241
Example 10-14 Conguring CoS-to-Queue Mapping on the Catalyst 3550

CAT3550(config-if)#wrr-queue cos-map 4 5
CAT3550(config-if)#wrr-queue cos-map 3 6 7
! Assigns CoS 5 to queue 4

! Reassigns CoS 6 and 7 to queue 3
Gigabit Ethernet ports can be congured to operate in a 4Q2T or 1P3Q2T manner, while
Catalyst 3550 Fast Ethernet ports can be congured to operate in a 4Q1T or 1P3Q1T
manner.
These drop thresholds can be congured in one of two ways: tail drop or WRED.
Tail drop is the default congestion-avoidance technique on Gigabit-capable Ethernet ports.
With tail drop, packets are queued until the thresholds are exceeded. Specically, all
packets with DSCPs assigned to the rst threshold are dropped until the threshold no longer
is exceeded. However, packets assigned to the second threshold continue to be queued and
sent as long as the second threshold is not exceeded.
You can modify the two tail-drop threshold percentages assigned to the four egress queues
by using the wrr-queue threshold interface conguration command. Each threshold value
is a percentage of the total number of allocated queue descriptors for the queue. The default
threshold is 100 percent for thresholds 1 and 2.
Alternately, you can enable WRED and congure the two threshold percentages assigned
to the four egress queues on a Gigabit-capable Ethernet port by using the wrr-queue
random-detect max-threshold interface conguration command. Each threshold percentage
represents where WRED starts to randomly drop packets. After a threshold is exceeded,
WRED randomly begins to drop packets assigned to this threshold. As the queue limit is
approached, WRED continues to drop more packets. When the queue limit is reached,
WRED drops all packets assigned to the threshold. By default, WRED is disabled.
If you use WRED thresholds, you cannot use tail drop, and vice versa. If WRED is disabled,
tail drop automatically is enabled with the previous conguration.
You modify the DSCP-to-threshold map to determine which DSCPs are mapped to which
threshold ID by using the wrr-queue dscp-map interface conguration command. By
default, all DSCPs are mapped to threshold 1; when this threshold is exceeded, all the
packets randomly are dropped.
In Example 10-15, WRED is enabled on the three remaining queues (because queue 4
previously was enabled as the expedite/strict-priority queue). The rst threshold for each
remaining queue is set to 40 percent, and the second threshold is set to 100 percent (the tail
of the queue). DSCP values of AF12, AF22, AF32, and AF42 (12, 20, 28, and 36, respectively)
are mapped to the rst thresholds. All other relevant DSCPs have been mapped to the
242
second threshold. Because the command allows only eight DSCP values to be mapped to a
threshold per command, a second entry is needed to map values DSCP values CS6 (48) and
CS7 (56) to the second threshold.
Example 10-15 Conguring WRED Thresholds and DSCP-to-Threshold Maps on a Catalyst 3550 Gigabit
Ethernet Interface
CAT3550(config-if)#wrr-queue
random-detect max-threshold
dscp-map 1 12 20 28 36
dscp-map 2 8 10 16 18 24 26
dscp-map 2 48 56
1 40 100
2 40 100
3 40 100
32 34
Catalyst 2970, 3650, and 3750

The Cisco Catalyst 2970 series switches are entry-level Gigabit Ethernet switches that
deliver wire-speed intelligent services for small and medium businesses and enterprise
branch ofces.
The Catalyst 2970 is available with Enhanced Image (EI) Cisco IOS Software, which
supports a rich QoS features set. Additionally, the 2970 supports identity-based network
services, enhancing the security of converged voice, video, and data networks, as in VoIP.
The Cisco Catalyst 3560 series switches is a line of xed conguration, enterprise-class,
IEEE 802.3af, and Cisco prestandard Power over Ethernet (PoE) switches. The Catalyst
3560 ts well in small enterprise wiring closets or branch ofce environments that are using
their LAN infrastructure for the deployment of IP phones, wireless access points, video surveillance, building management systems, and remote video kiosks. While the Catalyst 3560
supports IP Routing, it is better suited to the access layer (when L3 is deployed to the wiring
closet) than the distribution layer because its inline power capabilities are rarely required at
the distribution layer.
The Cisco Catalyst 3750 series switches represent the next step in the evolution of desktop
switches and feature Cisco StackWise technology. Cisco StackWise technology unites up
to nine individual Cisco Catalyst 3750 switches into a single logical unit, using special
stack-interconnect cables and stacking software (the stack interconnect is 32 Gbps). The
stack behaves as a single switching unit that is managed by a master switch elected from
one of the member switches. The master switch automatically creates and updates all the
switching and optional routing tables. A working stack can accept new members or delete
old ones without service interruption. Thus equipped, the Catalyst 3750 is optimized for
high-density Gigabit Ethernet deployments and is suitable for access and distribution layer
deployments, or even small-network core-layer requirements.
Catalyst 2970, 3650, and 3750
243
The Catalyst 3560 and 3750 series switches are available in the Standard Multilayer
Software Image (SMI) or the Enhanced Multilayer Software Image (EMI). The SMI feature
set includes all supported QoS tools and basic static and RIP routing functionality. The EMI
provides a richer set of enterprise-class features, including advanced hardware-based IP
unicast and multicast routing.
From a QoS perspective, the Catalyst 2970, 3560, and 3750 are identical and are considered
together in this discussion. The Catalyst 2970 and 3750 switches are shown in Figure 10-12.
Figure 10-12 Cisco Catalyst 2970 and 3750 Series Switches
By default, QoS is disabled on the 2970/3560/3750 platforms and must be enabled

explicitly with the global mls qos command.
Catalyst 2970/3560/3750 Classication, Marking, and Mapping

The Catalyst 2970/3560/3750 share many common QoS features and syntax with the
Catalyst 3550. Wherever feature functionality and syntax are identical to those of the 3550,
details and conguration examples are not repeated in this section.
Trust states and MQC/ACL classication and marking on the Catalyst 2970/3560/3750 are
identical in function and syntax to the 3550.
All Catalyst 2970/3560/3750 default CoS-to-DSCP, IP Precedence-to-DSCP, and DSCPto-CoS maps are identical to those for the 3550, as are mapping functions and syntax
(including DSCP mutation).
244
Catalyst 2970/3560/3750 Policing and Markdown

Policing and markdown functionality and syntax on the 2970/3560/3750 are identical to
those of the 3550, with the exception that the 2970/3560/3750 currently do not support perPort/per-VLAN policing.
The 2970/3560/3750 support 64 policers per interface. However, whereas the congurable
limit on the number of policers per interface is 64, the maximum number of policers per
port ASIC is 256. This limit present the options in the following list.
On a 24-port 10/100 + 2 Small Form-Factor Pluggable (SFP) modules chassis, there is

one port ASIC. This means that that the 256 policers must be shared among 26 ports,
with any 1 port having a maximum of 64 policers. However, if you congure four
ports with 64 policers each, there will be no more policers left for any other ports on
the chassis.
On a 48-port 10/100 + 4 SFP chassis, there are two port ASICs. This means that you
have two sets of 24 10/100 + 2 SFP ports internal to the switch. Each set of ports
supports a total of 256 policers. So, once again, the maximum policers on a port is 64,
but the maximum for each ASIC is 256.
These considerations should be kept in mind when designing for policing needs using the
Catalyst 2970/3560/3750 switch.
Catalyst 2970/3560/3750 Queuing and Dropping

Up to this point in the discussion, the Catalyst 2970, 3560 and 3750 series switches have
been shown to be remarkably identical to the Catalyst 3550 in terms of QoS features, functionality, and syntax. However, the platforms differ signicantly when it comes to queuing.
The rst major difference in queuing between the platforms is that the 2970/3560/3750
support ingress scheduling. However, ingress scheduling rarely is required because it
provides benets only if the (combined) input rates from any ports exceed the switching
fabric rate of the switch. This is extremely rare in most campus environments.
The Catalyst 2970/3560/3750 also supports four egress queues, one of which can be congured as an expedite/priority queue through the priority-queue out interface command.
Incidentally, on the Catalyst 2970/3560/3750, queue 1, not queue 4 (as on the 3550 and
2950) is used as an expedite queue (when congured).
Nonpriority queues are serviced through a shaped round-robin (SRR) algorithm that can be
congured to operate in one of two modes: shaped or sharing.
Actually, both the ingress and egress queues are serviced by SRR, which determines the
rate at which packets are sent. On the ingress queues, SRR sends packets to the stack ring.
On the egress queues, SRR sends packets to the egress interface. For ingress queues,
sharing is the default mode and is the only mode supported.
Catalyst 2970, 3650, and 3750
245
In shaped mode, the egress queues are guaranteed a percentage of the bandwidth and are
rate limited to that amount. Shaped trafc does not use more than the allocated bandwidth
even if the link is idle. Shaping provides a more even ow of trafc over time and reduces
the peaks and valleys of bursty trafc. With shaping, the absolute value of each weight is
used to compute the bandwidth available for the queues.
In shared mode, the queues share the bandwidth among them according to the congured
weights. The bandwidth is guaranteed at this level but is not limited to it. For example, if a
queue is empty and no longer requires a share of the link, the remaining queues can expand
into the unused bandwidth and can share it among them. With sharing, the ratio of the
weights determines the frequency of dequeuing; the absolute values are meaningless.
Additionally, both the ingress and egress queues use an enhanced version of the tail-drop
congestion-avoidance mechanism called weighted tail drop (WTD). As a frame is enqueued
to a particular queue, WTD uses the frames assigned CoS value or DSCP value to assign
it to different thresholds. If the threshold is exceeded for that CoS or DSCP value, the switch
drops the frame. Each queue supports (up to) three WTD thresholds: Two are congurable
(explicit WTD thresholds), and the third is noncongurable (implicit WTD threshold) and
is preset to the queue-full state (100 percent).
Thus, Catalyst 2970/3560/3750 switches can be congured to operate in 4Q3T mode or
1P3Q3T mode. The Catalyst 2970/3560/3750 also supports two queue sets, meaning that
different interfaces can be congured to operate in different queuing manners. For example,
some interfaces might be congured to operate in 4Q3T, and others might be congured to
operate in 1P3Q3T, depending on which queue set ID (qset-id) they are assigned to.
The allocated memory assigned to each queue can be modied (from the default 25 percent
per queue) by using the mls qos queue-set output qset-id buffers allocation1 ... allocation4
global conguration command. The sum of all the allocated buffers represents the reserved
pool, and the remaining buffers are part of the common pool.
The Catalyst 2970/3560/3750 switches use a buffer-allocation scheme to reserve a minimum number of buffers for each egress queue, to prevent any queue or port from consuming
all the buffers and depriving other queues, and to determine whether to grant buffer space
to a requesting queue. The switch determines whether the target queue has not consumed
more buffers than its reserved amount (under limit), whether it has consumed all of its maximum buffers (over limit), or whether the common pool is empty (no free buffers) or not
empty (free buffers). If the queue is not over limit, the switch can allocate buffer space from
the reserved pool or from the common pool (if it is not empty). If there are no free buffers
in the common pool, or if the queue is over limit, the switch drops the frame.
You guarantee the availability of buffers, set drop thresholds, and congure the maximum
memory allocation for a queue set by using the following global conguration command:
mls qos queue-set output qset-id threshold queue-id drop-threshold1
drop-threshold2 reserved-threshold maximum-threshold
Each threshold value is a percentage of the queues allocated memory.
246
In Example 10-16, queue set 1 has buffers allotted among queues 1 through 4 by the percentages 35, 30, 25, and 10. If 400 buffers are available, this would translate to an allocation
of 140, 120, 100, and 40 buffers to queues 1 through 4, respectively. Also, WTD thresholds
on queues 2 through 4 have been set to 40 percent of the buffer depth and 100 percent of
the buffer depths. Each queue has its buffer allocation fully reserved (100 percent) and has
its maximum memory threshold also set to 100 percent. Additionally, each queue has its
reserved threshold (an amount of memory to be guaranteed or reserved for the queuethe
range is 1 to 100 percent) set to 100 percent. Furthermore, each queue has its maximum
threshold (which enables a queue in the full condition to obtain more buffers than are
reserved for it and is the maximum memory that the queue can have before the packets are
droppedthe range is 1 to 400 percent) set to 100 percent.
Finally, Fast Ethernet 0/1 on the switch that has been designated as stack member 2 is
assigned to queue set 1, and queue 1 (of queue set 1) is dened as an expedite/priority
queue.
Example 10-16 Conguring Per-Queue Buffer Allocations and Thresholds on a Catalyst 2970/3560/3750
CAT3750(config)#mls qos queue-set output 1 buffers 35 30 25 10
CAT3750(config)#mls qos queue-set output 1 threshold 2 40 100 100 100
CAT3750(config)#int fa 2/0/1
CAT3750(config-if)#queue-set
queue-set 1
CAT3750(config-if)#priority-queue out
Similar to WRR, SRR allows the weights assigned to each queue to be modied. The ratio
of these weights determines the frequency by which the SRR scheduler sends packets out
from each queue. Shaping weights smooth or rate limit bursty or domineering ows. The
inverse ratio (1/weight) determines the shaping bandwidth for each queue. For example, if
a queue is to be limited to 10 percent, the shaping weight for that queue is set to 10. A
shaping weight of 0 means that the queue is operating in sharing mode. Shaped mode
overrides shared mode.
In shared mode, the queues share the bandwidth among them according to the congured
weights. The bandwidth is guaranteed at this level but is not limited to it. For example, if a
queue empties and does not require a share of the link, the remaining queues can expand
into the unused bandwidth and share it among them. With sharing, the ratio of the weights
determines the frequency of dequeuing; the absolute values are meaningless.
If an expedite queue has been congured, this strict-priority designation overrides any SRR
weights for queue 1.
In Example 10-17, the queues are assigned 35 percent, 30 percent, 25 percent, and 10
percent shared bandwidth allocations. But, in actuality, queue 1s bandwidth allocation is
ignored because it is congured as a strict-priority queue. Queues 2 and 3 are operating in
shared mode (because their shaped weight is set to 0). Queue 4 is operating in shaped mode
(shaped weight overrides shared weight) and is limited to no more than 10 percent of the
bandwidth, regardless of whether more is available.
Catalyst 4500
247
Example 10-17 Conguring SRR Shaping and Sharing Weights on a Catalyst 2970/3560/3750
CAT3750(config)#int gig 1/0/4
CAT3750(config-if)#srr-queue bandwidth share 35 30 25 10
CAT3750(config-if)#srr queue bandwidth shape 0 0 0 10
CAT3750(config-if)#queue-set 1
CAT3750(config-if)#priority-queue out
Finally, the Catalyst 2970/3560/3750 switches allow packets to be assigned to queues either
by CoS values or by DSCP values, using CoS-to-queue/threshold maps or DSCP-to-queue/
threshold maps.
In Example 10-18, DSCP EF (46) is assigned to queue 1 (the expedite queue). DSCP CS7
(56) and CS6 (48) are assigned to queue 2 threshold 2, and DSCP AF31 (26) and DSCP
CS3 are assigned to queue 2 threshold 1. DSCP 0 is assigned to queue 3. Bulk (DSCP
AF1110) is assigned to queue 4 threshold 2, while Scavenger (DSCP CS18) trafc is
assigned to queue 4 threshold 1.
Example 10-18 Assigning DSCP Values to Queues/Thresholds on a Catalyst 2970/3560/3750
CAT3750(config)#mls
CAT3750(config)#mls
CAT3750(config)#mls
CAT3750(config)#mls
CAT3750(config)#mls
CAT3750(config)#mls
qos
qos
qos
qos
qos
qos
srr-queue
srr-queue
srr-queue
srr-queue
srr-queue
srr-queue
output
output
output
output
output
output
dscp-map
dscp-map
dscp-map
dscp-map
dscp-map
dscp-map
queue
queue
queue
queue
queue
queue
1
2
2
3
4
4
threshold
threshold
threshold
threshold
threshold
threshold
1
2
1
2
2
1
46
48 56
24 26
0
10
8
Catalyst 4500
The Cisco Catalyst 4500, shown in Figure 10-13, nds its niche between desktop switches
and the agship Catalyst 6500. As such, it can be found in the access layer, the distribution
layer, and even the core layer in midsize networks. Service providers also utilize the
Catalyst 4500 because of its support of metro Ethernet options combined with its resiliency/
failover options.
248
QoS support for the Catalyst 4000 Supervisor I and II cards (running CatOS) was quite
limited and, as such, is omitted from this discussion. However, the Catalyst 4500
Supervisors II+ through V (running Cisco IOS) boast a much richer QoS toolset and are
examined further.
Many of the QoS commands on the 4500 are similar to those on the switches previously
discussed, with the exception of omitting the mls keyword at the beginning of the
statement.
For example, by default, QoS is disabled on the 4500 and must be enabled with the global
command qos (whereas, on previously discussed platforms, this command would have
been mls qos). When QoS is disabled, the port interface trust states default to Trust DSCP.
When QoS is enabled, this trust state changes to Untrusted, and (with no other QoS parameters explicitly dened) all egress trafc is marked to CoS 0 and DSCP 0. Incidentally,
QoS can be disabled on a per-interface basis using the interface command no qos (see
Example 10-19).
Example 10-19 Enabling QoS on a Catalyst 4500
CAT4500(config)#qos
qos
QoS policies can be applied to individual ports or to VLANs. MQC policies can be applied
to VLANs by attaching the service-policy statement to the VLAN interface and conguring
all ports belonging to the VLAN to use VLAN-based QoS (using the qos vlan-based
interface command). If the interface is congured to use VLAN-based QoS, the trafc
received or sent through the interface is classied, policed, and marked according to the
policy map attached to the VLAN (congured on the VLAN interface) to which the packet
belongs. If no policy map is attached to the VLAN to which the packet belongs, the policy
map attached to the interface is used. Furthermore, VLAN-based policies supercede any
policies congured on individual ports if the port has been assigned to operate in VLANbased QoS mode.

The Catalyst 4500 can be congured to trust CoS or DSCP. Additionally, it can be
congured to selectively trust a device, such as an IP phone (through CDP detection).
Conguring trusts is very similar to previously discussed syntaxes, but without the mls
keyword. Refer to Example 10-20.
Example 10-20 Enabling QoS on a Catalyst 4500
CAT4500(config)#interface range fa 2/1 - 48
CAT4500(config-if-range)#qos
qos trust cos
CAT4500(config-if-range)#qos
qos trust device cisco-phone
Catalyst 4500
249
The conguration of MQC/ACLbased classication is the same as in previously discussed

examples. However, if the set command is used in a policy map for marking, IP routing
(disabled by default) must be enabled on the 4500. At a minimum, this requires enabling IP
routing and setting a default route to a next-hop device capable of forwarding.
CoS-to-DSCP maps are the same as on all Catalyst switches (shown in Table 10-1). DSCPto-CoS maps are identical to the Cisco 3550 and 3750 DSCP-to-CoS maps (shown in
Table 10-3) because the 4500 also supports the full DSCP range of values (0 to 63).
As always, these mapping values can be modied away from the default settings. The
syntax is again similar, without the mls keyword.
As shown in Example 10-21, CoS 5 is mapped to DSCP 46 (EF), and DSCP CS1 (8) is
mapped to CoS 0. In the rst map (CoS-to-DSCP), eight separate values must be supplied
(in order) that are to correspond to the CoS values of 0 through 7. Notice that the sixth value
(corresponding to CoS 5) is modied away from the default (40) and explicitly is set to
46 (DSCP EF). With the second mapping, up to eight DSCP values (separated by spaces)
can be mapped to a CoS value, which follows the to keyword.
Example 10-21 Conguring Mapping Modications on the Catalyst 4500
CAT4500(config)#qos map cos-dscp 0 8 16 24 32 46 48 56
CAT4500(config)#qos map dscp-cos 8 to 0
DSCP mutation currently is not supported on the Catalyst 4500.

The Catalyst 4500 supports individual and aggregate policers, as well as port-based and
VLAN-based policing, in both the ingress and egress directions.
The switch limits the number of input policers and output policers to 1024 each. But in
actuality, because the software reserves four input policers and four output policers for null
processing, this leaves 1020 policers available for input policing and 1020 policers
available for output processing.
The Catalyst 4500 has a handy syntax for dening bit rates: It uses the prexes k for kilo,
m for mega, and g for giga. Instead of getting lost in zeros, a value such as 1,250,000,000 bps
can be expressed as 1.25 Gbps. This syntax alternative can reduce conguration errors
introduced by typos.
In Example 10-22, individual Bulk Data trafc ows (marked on a data center server
connected to Fast Ethernet 2/12) in excess of 10 Mbps are marked down from AF11 (10)
to AF12 (12).
250
Example 10-22 Conguring an Individual Policer on a Catalyst 4500

CAT4500(config)#qos map dscp policed 10 to dscp 12
CAT4500(config-cmap)# match ip dscp af11
CAT4500(config-cmap)#policy-map BULK-MARKDOWN
CAT4500(config-pmap)# class BULK
CAT4500(config-pmap-c)#police 10 mbps 8000 conform-action transmit
exceed-action policed-dscp-transmit
CAT4500(config-pmap-c)#int fa 2/12
CAT4500(config-if)#service-policy input BULK-MARKDOWN
In addition to individual policers, an aggregate policer can be applied to a Catalyst 4500

access-to-distribution uplink to mark down cumulatively out-of-prole trafc.
Building on the previous example, all Bulk Data trafc in excess of 100 Mbps (regardless
of how many ows are involved) is marked down to AF12 on the Gigabit 1/1 uplink. The
policed-DSCP map does not need to be redened, but both individual and aggregate
policers can leverage it. Refer to Figure 10-23.
Example 10-23 Conguring an Aggregate Policer on a Catalyst 4500
CAT4500(config)#qos aggregate-policer BULK-MARKDOWN 100 mbps
8000 conform transmit exceed-action policed-dscp-transmit
CAT4500(config)#policy-map BULK-MARKDOWN-AGG
CAT4500(config-pmap-c)# police aggregate BULK-MARKDOWN-AGG
CAT4500(config-pmap-c)#int gig 1/1
CAT4500(config-if)# service-policy output BULK-MARKDOWN-AGG

The Catalyst 4500 supports four egress queues, one of which can be congured as a priority
queue. No thresholds (other than the tail of the queue) are congurable (at the time of
writing), allowing the 4500 to be congured through a 4Q1T or 1P3Q1T model.
On the Catalyst 4500, Gigabit Ethernet interfaces employ a queue size of 1920 packets, and
Fast Ethernet interfaces use a queue size of 240 packets.
Only queue 3 can be congured as a priority queue.
By default, nonpriority queues are scheduled in a round-robin manner and share bandwidth
equally among them. However, bandwidth allocations can be congured explicitly on the
following ports:
Uplink ports on supervisor engines

Ports on the WS-X4306-GB line card
The two 1000BASE-X ports on the WS-X4232-GB-RJ line card
Catalyst 4500
251
The rst two ports on the WS-X4418-GB line card

The two 1000BASE-X ports on the WS-X4412-2GB-TX line card
Minimum bandwidth allocations can be dened in absolute kbps (again, the k, m, and g
prexes can be used to dene the rate in bits per second). Alternatively, minimum bandwidth allocations also can be expressed as percentages.
Complementarily, each transmit queue can be congured to transmit a maximum rate using
the shape command. This command enables you to specify the maximum rate of trafc that
the queue services. Any trafc that exceeds the congured shape rate is queued and transmitted at the congured rate. If the burst of trafc exceeds the size of the queue, packets are
dropped to maintain transmission at the congured shape rate. As with the bandwidth command, the shaped rate can be expressed in absolute bits per second (the k, m, and g prexes
can be used to dene the rate in bits per second) or as a percentage.
In Example 10-24, queue 3 is assigned to be a strict-priority queue. Queue 1 is assigned
30 percent of the available bandwidth (after the priority queue has been fully serviced), and
queue 2 is assigned 25 percent. Queue 4 is assigned to be shaped to 10 percent.
Example 10-24 Conguring a Priority Queue and Bandwidth Allocations on a Catalyst 4500
CAT4500(config-if)#tx-queue
tx-queue 1
CAT4500(config-if-tx-queue)#bandwidth
CAT4500(config-if-tx-queue)#tx-queue
tx-queue 2
CAT4500(config-if-tx-queue)#bandwidth
CAT4500(config-if-range)#tx-queue
tx-queue 3
CAT4500(config-if-tx-queue)#priority
priority high
CAT4500(config-if-tx-queue)#tx-queue
tx-queue 4
CAT4500(config-if-tx-queue)#shape
shape percent 10
The Catalyst 4500 supports DSCP-to-queue maps, with the default DSCP-to-queue
assignment shown in Table 10-4.
Table 10-4
Catalyst 4500 Default DSCP-to-Queue Maps

DSCP Value
Transmit Queue
DSCP 0-15
Queue 1
DSCP 16-31
Queue 2
DSCP 32-47
Queue 3
DSCP 48-63
Queue 4
These assignments can be modied with the qos map dscp dscp to tx-queue queue command. Up to eight DSCP values (separated by spaces) can be assigned to a queue in a single
command.
252
Continuing from Example 1-24, queue 1 carries Mission-Critical trafc, such as NetworkControl (CS7/56), Routing (CS6/48), Call-Signaling (AF31/26), Transactional Data (AF21/18),
and Network-Management (CS2/16). Voice (EF/46) is mapped to priority queue 3. Queue 2
is assigned as a Best-Effort queue (DSCP 0). Less-than-best-effort trafc, such as Bulk
Data (AF11/AF12, which are DSCP 10 and 12, respectively) and Scavenger (CS1/8) trafc,
is assigned to queue 4.
Example 10-25 Conguring DSCP-to-Queue Assignments on a Catalyst 4500
CAT4500(config)#qos
CAT4500(config)#qos
CAT4500(config)#qos
CAT4500(config)#qos
map
map
map
map
dscp
dscp
dscp
dscp
56 48 26 18 16 to tx-queue 1
0 to tx-queue 2
46 to tx-queue 3
8 10 12 to tx-queue 4
As previously noted, the Catalyst 4500 does not (yet) support drop thresholds, but it does
support a congestion-avoidance feature called dynamic buffer limiting (DBL).
DBL tracks the queue length for each trafc ow in the switch. When the queue length of a
ow exceeds its limit, DBL drops packets or sets the explicit congestion notication (ECN)
bits in the packet headers. ECN bits, based on RFC 3168, were covered in Chapter 6,
Congestion-Avoidance Tools.
DBL classies ows in two categories, adaptive and aggressive. Adaptive ows reduce the
rate of packet transmission after it receives congestion notication. Aggressive ows do not
take corrective action in response to congestion notication.
Queue length is measured by the number of packets. The number of packets in the queue
determines the amount of buffer space that a ow is given. When a ow has a high queue
length, the computed value is lowered. This enables new incoming ows to receive buffer
space in the queue. This allows all ows to get a proportional share of packets through the
queue.
DBL, with ECN notication, can be enabled globally using the command in Example 10-26.
Example 10-26 Conguring DBL with ECN on a Catalyst 4500
CAT4500(config)#qos
qos dbl exceed-action ecn
Catalyst 6500
The Catalyst 6500, shown in Figure 10-14, is indisputably Ciscos main switch and, as
such, boasts the richest feature sets, not only in QoS, but in all services, such as high
availability, multicast, security, and management. It is the preferred switch in large
enterprise environments at all layers: access, distribution, and core. Additionally, the
Catalyst 6500 is very prevalent in service provider environments. In the context of our
Catalyst 6500
253
discussions, Catalyst 6500 is used to refer to Catalyst 6500s with Supervisor 2 (PFC2)
and Catalyst 6500s with Supervisor 720 (PFC3) because these are the current versions of
supervisor hardware at the time of writing.
Hardware QoS on the Catalyst 6500 is performed principally within the Policy Feature
Card (PFC), which is now in its third generation (with each successive generation of PFC
having progressively richer QoS featuresrefer to Cisco Catalyst 6500 documentation for
complete details).
The PFC is responsible for classication, marking, mapping, and policing. Individual line
cards perform the queuing and dropping functions. Thus, multiple combinations of queues
and thresholds (varying by line card) can exist for 6500 switches.
On the Catalyst 6500, PFC QoS can be congured in one of two ways:
NOTE
Hybrid modeThis mode uses the Catalyst Operating System (CatOS) to congure
the Supervisor and PFC. The Layer 3 forwarding engine (the Multilayer Switch
Feature Card, or MSFC) then is congured using Cisco IOS. The term hybrid is used
to convey that two operating systems (and two independent consoles) are used to
manage the Catalyst 6500.
Native modeThis mode uses a single image of Cisco IOS to congure the
Supervisor + PFC, along with the MSFC. Only a single console is required to manage
the Catalyst 6500.
For the QoS features discussed in this chapter, both the CatOS and Cisco IOS commands
are shown in the examples.
In both cases, QoS is disabled by default and needs to be globally enabled, as shown in
Example 10-27.
254
Example 10-27 Enabling QoS Globally on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set qos enable
QoS is enabled.
CAT6500-CATOS> (enable)
CAT6500-IOS(config)#mls
mls qos
CAT6500-IOS(config)#
QoS policies on the 6500 can be applied at the individual port level or on a per-VLAN basis
(but, by default, all ports are congured for port-based QoS). This implies that all classication
policies, and marking and policing, congured for the port are applicable only to the port
to which they are applied. Conversely, VLAN-based QoS enables the administrator to apply
a uniform policy to all ports congured for a specic VLAN. VLAN-based QoS might be
preferred in certain situations, such as when voice VLANs have been deployed for IP
Telephony environments. To alter the port policy from the default port-based QoS setting
to VLAN-based QoS, use the commands shown in Example 10-28.
Example 10-28 Conguring VLAN-Based QoS on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set port qos 3/1 vlan-based
Qos interface is set to vlan-based for ports 3/1
CAT6500-IOS(config)#interface FastEthernet3/1
CAT6500-IOS(config-if)#mls
mls qos vlan-based
When VLAN-based QoS has been enabled, service-policy statements are attached to the
VLAN interfaces, which then apply to all ports assigned to the VLAN.

All Catalyst 6500 line cards support the trusting of CoS markings. However, not all of them
support DSCP trust or IP Precedence trust (refer to Cisco Catalyst 6500 documentation for
complete details).
By default, when QoS is enabled globally on the 6500, the trust state of all ingress ports
is untrusted. In Example 10-29, the default trust state is changed from Untrusted to Trust
CoS.
Catalyst 6500
255
Example 10-29 Conguring Trust on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set port qos 3/1 trust trust-cos
Port 3/1 qos set to trust-cos
CAT6500-IOS(config)#interface FastEthernet3/1
CAT6500-IOS(config-if)#mls
mls qos trust cos
CAT6500-IOS(config-if)#
NOTE
Under some circumstances, the administrator must perform additional conguration steps
beyond what is noted in the preceding commands. For example, the WS-X6224/6248 and
WS-X6324/6348 line cards have hardware restrictions that prevent them from passing CoS
values to the switching engine. Although the line cards recognize the arriving CoS and use
that value for input scheduling, when the frame header is passed to the switching engine for
forwarding, the CoS value for the frame is not preserved and is rewritten to 0. Therefore, it
is necessary to congure additional commands, as shown in Example 10-30.
Example 10-30 Alternate Trust Conguration on the Catalyst 6500 (CatOS)

CAT6500-CATOS> (enable) set port qos 3/1 trust trust-cos
Trust type trust-cos not supported on this port.
Receive thresholds are enabled on port 3/1.
Port 3/1 qos set to untrusted.
CAT6500-CATOS> (enable) set qos acl ip TRUSTCOS trust-cos any
Warning: ACL trust-cos should only be used with ports that are also
configured with port trust=trust-cos.
TRUSTCOS editbuffer modified. Use 'commit' command to apply changes.
CAT6500-CATOS> (enable) commit qos acl TRUSTCOS
QoS ACL 'TRUSTCOS' successfully committed.
CAT6500-CATOS> (enable) set qos acl map TRUSTCOS 3/1
ACL TRUSTCOS is successfully mapped to port 3/1.
The old ACL mapping is replaced by the new one.
The PFC also can be congured to identify and mark ows to specic DSCP values by
either Layer 3 or Layer 4 parameters, and dened in access lists. This is shown in Example 10-31, in which TCP trafc within the destination port range of 9000 to 9005 is marked
as Transactional Data (DSCP AF21 or 18).
256
Example 10-31 ACL-Based Classication and Marking on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set qos acl ip TRANSACTIONAL-DATA dscp 18
tcp any any range 9000 9005
TRANSACTIONAL-DATA editbuffer modified. Use 'commit' command to apply changes.
CAT6500-CATOS> (enable) commit qos acl TRANSACTIONAL-DATA
QoS ACL 'TRANSACTIONAL-DATA' successfully committed.
CAT6500-CATOS> (enable) set qos acl map TRANSACTIONAL-DATA 3/1
ACL TRANSACTIONAL-DATA is successfully mapped to port 3/1.
The old ACL mapping is replaced by the new one.
CAT6500-IOS(config)#access-list
access-list 100 permit tcp any any range 9000 9005
CAT6500-IOS(config)#class-map TRANSACTIONAL-DATA
CAT6500-IOS(config-cmap)#match
match access-group 100
CAT6500-IOS(config-cmap)#policy-map ACCESS-MARKING
CAT6500-IOS(config-pmap)#class TRANSACTIONAL-DATA
CAT6500-IOS(config-pmap-c)#set
set dscp af21
CAT6500-IOS(config-pmap-c)#interface FastEthernet 3/1
CAT6500-IOS(config-if)#service-policy
service-policy input ACCESS-MARKING
The default CoS-to-DSCP maps, IP Precedence-to-DSCP maps, and DSCP-to-CoS maps

are the same as have been discussed on previous switches (except for the Catalyst 2950,
which does not support the full DSCP range at the time of writing).
CoS-to-DSCP maps can be overwritten from the defaults using the syntaxes shown. In both
cases, eight DSCP values must be supplied that correspond to CoS values 0 through 7,
respectively. In Example 10-32, CoS 5 is mapped to DSCP EF (46), CoS 4 is mapped to
DSCP AF41 (34), and CoS 3 is mapped to DSCP AF31 (26).
Example 10-32 CoS-to-DSCP Mapping Modication on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set qos cos-dscp 0 8 16 26 34 46 48 56
QoS cos-dscp-map set successfully.
CAT6500-IOS(config)#mls qos map cos-dscp 0 8 16 26 34 46 48 56
IP Precedence-to-DSCP maps use identical syntax, except that the cos keyword is replaced
with ipprec (CatOS) and ip-prec (Cisco IOS).
Default DSCP-to-CoS modications can be congured, as shown in Example 10-33. In this
example, DSCP values CS1 (8) through DSCP 15 are remapped to CoS 0.
Catalyst 6500
257
Example 10-33 DSCP-to-CoS Mapping Modication on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set qos dscp-cos-map 8-15:0
QoS dscp-cos-map set successfully.
CAT6500-IOS(config)#mls qos map dscp-cos 8 9 10 11 12 13 14 15 to 0
DSCP-to-DSCP maps, or DSCP mutation maps, also can be congured explicitly on the
Catalyst 6500, but DSCP mutation currently is in native IOS mode only. Up to 15 DSCP
mutation maps can be congured on the Catalyst 6500. The commands to congure
DSCP mutation maps on the 6500 are identical to those used in the examples of DSCP
mutation previously discussed in this chapter.

The Catalyst 6500 supports two main types of policers: microow and aggregate. Both
microow and aggregate policers can be congured on a per-port or per-VLAN basis. The
Catalyst 6500 supports 63 microow and 1023 aggregate policers.
Microow policers differ from individual policers (which limit trafc on a per-interface
basis), in that they limit trafc on a per-ow basis. A ow is dened by a socket (source/
destination address pairing combined with an identical Layer 4 protocol and the
source/destination port pairing). However, microow policers can be congured to use
only the source or only the destination addresses for policing purposes.
NOTE
By default, microow policers affect only trafc routed by the MSFC. To enable microow
policing of bridged trafc, either the CatOS set qos bridged-microow-policing enable
global command or the Cisco IOS mls qos bridged interface command must be used.
The commands to congure microow policing on the Catalyst 6500 are shown in
Example 10-34, where no single ow is permitted to exceed 1 Mbps. In CatOS, this
requires a microow policer to be dened and referenced by an ACL. The ACL, in turn, is
committed and mapped to the interface. In Cisco IOS, an MQC policy is dened with the
police ow keywords and is attached to an interface. Notice in Example 10-34 that policing
rates might be adjusted marginally because of hardware granularity restrictions.
Aggregate policers can be dened on a per-interface basis. Alternatively, you can dene
named aggregate policers that can be applied to multiple interfaces.
258
Example 10-34 Conguring Microow Policing on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set qos policer microflow ONE-MEG-FLOWS
rate 1000 burst 8 drop
QoS policer for microflow ONE-MEG-FLOWS created successfully.
Rate is set to 992 and burst is set to 8 in hardware due to hardware granularity.
CAT6500-CATOS> (enable) set qos acl ip ONE-MEG-FLOW-ACL dscp 0
microflow ONE-MEG-FLOWS ip any any
ONE-MEG-FLOW-ACL editbuffer modified. Use 'commit' command to apply changes.
CAT6500-CATOS> (enable) commit qos acl ONE-MEG-FLOW-ACL
QoS ACL 'ONE-MEG-FLOW-ACL' successfully committed.
CAT6500-CATOS> (enable) set qos acl map ONE-MEG-FLOW-ACL 3/1
ACL ONE-MEG-FLOW-ACL is successfully mapped to port 3/1.
CAT6500-IOS(config)#access-list 1 permit any
CAT6500-IOS(config)#class-map ANY
CAT6500-IOS(config-cmap)#match access-group 1
CAT6500-IOS(config-cmap)#policy-map MICROFLOW
CAT6500-IOS(config-pmap)#class ANY
CAT6500-IOS(config-pmap-c)#police
police flow 1000000 8000 conform transmit
exceed-action drop
CAT6500-IOS(config-pmap-c)#interface FastEthernet 3/1
CAT6500-IOS(config-if)#service-policy input MICROFLOW
Furthermore, aggregate policers on the PFC2 or PFC3 support the added functionality of
dual-rate policing, as dened in RFC 2698 and overviewed in Chapter 4, Policing and
Shaping Tools, under the Two-Rate Three-Color Marker section. Dual-rate policers
require dening not only a CIR, but also a PIR. However, the out-of-prole PIR action
(violate) cannot be less severe than the out-of-prole CIR action (exceed). For instance,
if the exceed action is to mark down, the violate action cannot be to transmit.
In Example 10-35, FTP trafc is transmitted with DSCP AF11 (DSCP 10) if it is less than
1 Mbps. Any FTP trafc over 1 Mbps but less than 2 Mbps is marked down to AF12 (DSCP
12). Any FTP trafc in excess of 2 Mbps is marked down further to AF13 (DSCP 14). The
aggregate policing policy is applied to VLAN 10 as a whole.
Example 10-35 Conguring Dual-Rate Aggregate Policers on the Catalyst 6500 (CatOS and
Cisco IOS)
CAT6500-CATOS> (enable) set qos policed-dscp-map normal 10:12
CAT6500-CATOS> (enable) set qos policed-dscp-map excess 10:14
CAT6500-CATOS> (enable) set qos policer aggregate FTP-POLICER
rate 1000 policed-dscp erate 2000 policed-dscp burst 8 eburst 8
QoS policer for aggregate FTP-POLICER created successfully.
Rate is set to 992 and erate is set 1984 and burst is set to 8
and eburst is set to 8 in hardware due to hardware granularity.
CAT6500-CATOS> (enable) set qos acl ip FTP-ACL dscp 10 aggregate FTP-POLICER
tcp any any eq ftp
FTP-ACL editbuffer modified. Use 'commit' command to apply changes.
Catalyst 6500
259
Example 10-35 Conguring Dual-Rate Aggregate Policers on the Catalyst 6500 (CatOS and
Cisco IOS) (Continued)

CAT6500-CATOS> (enable) set qos acl ip FTP-ACL dscp 10 aggregate FTP-POLICER
tcp any any eq ftp-data
FTP-ACL editbuffer modified. Use 'commit' command to apply changes.
CAT6500-CATOS> (enable) commit qos acl FTP-ACL
QoS ACL 'FTP-ACL' successfully committed.
CAT6500-CATOS> (enable) set qos acl map FTP-ACL 10
ACL FTP-ACL is successfully mapped to vlan 10.
mls qos map policed-dscp normal-burst 10 to 12
mls qos map policed-dscp max-burst 10 to 14
mls qos aggregate-policer FTP-AGG-POLICER 1000000 pir 2000000
conform-action set-dscp-transmit 10 exceed-action policed-dscp-transmit
violate-action policed-dscp-transmit
CAT6500-IOS(config)#access-list 100 permit tcp any any eq ftp
CAT6500-IOS(config)#access-list 100 permit tcp any any eq ftp-data
CAT6500-IOS(config)#class-map FTP
CAT6500-IOS(config-cmap)#match access-group 100
CAT6500-IOS(config-cmap)#policy-map FTP-MQC-POLICER
CAT6500-IOS(config-pmap)#class FTP
CAT6500-IOS(config-pmap-c)#police
police aggregate FTP-AGG-POLICER
CAT6500-IOS(config-pmap-c)#interface vlan 10
CAT6500-IOS(config-if)#service-policy
service-policy input FTP-MQC-POLICER
Microow and aggregate policers are not mutually exclusive. For example, you could
create a microow policer with a bandwidth limit suitable for individuals in a group, and
you could create a named aggregate policer with bandwidth limits suitable for the group as
a whole. You could include both policers in policy map classes that match the groups
trafc. The combination would affect individual ows separately and the group aggregately.

As mentioned, queuing and threshold support on Catalyst 6500 varies from line card to line
card. To check the queuing structure of the line card or port in question, use the commands
in Example 10-36.
Example 10-36 Queue Structure Verication on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) show port capabilities 1/1
Model
WS-X6K-SUP1A-2GE
Port
1/1
Type
1000BaseSX
Speed
1000
continues
260
Example 10-36 Queue Structure Verication on the Catalyst 6500 (CatOS and Cisco IOS) (Continued)
Duplex
Trunk encap type
Trunk mode
Channel
Broadcast suppression
Flow control
Security
Dot1x
Membership
Fast start
QOS scheduling
full
802.1Q,ISL
on,off,desirable,auto,nonegotiate
yes
percentage(0-100)
receive-(off,on,desired),send-(off,on,desired)
yes
yes
static,dynamic
yes
rx-(1p1q4t),tx-(1p2q2t)
1p2q2t
CAT6500-IOS#show
show queueing interface gig 1/1
Interface GigabitEthernet1/1 queueing strategy:
Port QoS is enabled
Trust state: trust DSCP
Extend trust state: not trusted [COS = 0]
Default COS is 0
Queueing Mode In Tx direction: mode-cos
Transmit queues [type = 1p2q2t
1p2q2t]:
...
CAT6500-IOS#
Weighted Round-Robin
The commands display some of the following queue structures:
2q2t indicates two standard queues, each with two congurable tail-drop thresholds.
1p2q1t indicates the following:
One strict-priority queue
Two standard queues, each with one WRED-drop threshold
One noncongurable (100 percent) tail-drop threshold

Two standard queues, each with two congurable WRED-drop thresholds

Three standard queues, each with one threshold congurable as either
WRED drop or tail drop
One noncongurable (100 percent) tail-drop threshold
Catalyst 6500
261

Three standard queues, each with eight thresholds, with each threshold
congurable as either WRED drop or tail drop

Seven standard queues, each with eight thresholds, with each threshold
congurable as either WRED drop or tail drop
For port types with a strict-priority queue, the switch services trafc in the strict-priority
transmit queue before servicing the standard queues. When the switch is servicing a
standard queue, after transmitting a packet, it checks for trafc in the strict-priority queue.
If the switch detects trafc in the strict-priority queue, it suspends its service of the standard
queue and completes service of all trafc in the strict-priority queue before returning to the
standard queue.
Catalyst 6500 PFC QoS schedules trafc through the transmit queues based on Layer 2 CoS
values. In the default conguration, PFC QoS assigns all trafc with CoS 5 to the strictpriority queue (if present); PFC QoS assigns all other trafc to standard queues. In the
absence of a strict-priority queue, PFC QoS assigns all trafc to the standard queues.
NOTE
The default queue sizes, thresholds denitions, and bandwidth allocation ratios are usually
adequate for most cases. In the rare event that these need to be tuned, experienced administrators should refer to Cisco Catalyst 6500 documentation for default values for threshold
percentages, queue sizes, and bandwidth-allocation ratios for each queuing structure.
Keep these points in mind when conguring CoS to queue/threshold assignments and
parameters:
Number 1 is the lowest-priority standard queue.

Higher-numbered queues are higher-priority standard queues.
When you congure multiple-threshold standard queues, note the following:
The rst percentage that you enter sets the lowest-priority threshold.
Always set highest-numbered threshold to 100 percent.
The second percentage that you enter sets the next highest-priority threshold.
The last percentage that you enter sets the highest-priority threshold.
The percentages range from 1 to 100. A value of 10 indicates a threshold when the
buffer is 10 percent full.
262
When conguring the WRED-drop thresholds, note the following:
Each WRED-drop threshold has a low-WRED and a high-WRED value.
The low-WRED value is the trafc level under which no trafc is dropped. The lowWRED value must be lower than the high-WRED value.
The high-WRED value is the trafc level above which all trafc is dropped.
Low-WRED and high-WRED values are a percentage of the queue capacity (the
range is from 1 to 100).
Trafc in the queue between the low- and high-WRED values has an increasing
chance of being dropped as the queue lls.
In the 1P2Q2T model shown in Example 10-37, Scavenger trafc (CoS 1) is mapped to the
rst standard queue, rst threshold; Best-Effort trafc (CoS 0) is mapped to the rst queue,
second threshold; Transactional Data, Network-Management trafc (CoS 2), and Video
(CoS 4) are mapped to the second queue, rst threshold; Call-Signaling and NetworkControl trafc (CoS values 3, 6, and 7) are mapped to the second queue, second threshold.
Voice (CoS 5) is mapped to the priority queue.
Example 10-37 Conguring CoS-to-Queue/Threshold Assignments on the Catalyst 6500 (CatOS and Cisco IOS)
CAT6500-CATOS> (enable) set qos map 1p2q2t tx 1 1 cos 1
QoS tx priority queue and threshold mapped to cos successfully.
CAT6500-CATOS> (enable) set qos map 1p2q2t tx 2 1 cos 2,4
CAT6500-CATOS> (enable) set qos map 1p2q2t tx 2 2 cos 3,6,7
CAT6500-IOS(config)# interface GigabitEthernet2/1
CAT6500-IOS(config-if)#wrr-queue cos-map 1 1 1
cos-map configured on: Gi2/1 Gi2/2 Gi2/3 Gi2/4 Gi2/5
CAT6500-IOS(config-if)#wrr-queue cos-map 1 2 0
CAT6500-IOS(config-if)#wrr-queue cos-map 2 1 2 4
CAT6500-IOS(config-if)#wrr-queue cos-map 2 2 3 6 7
CAT6500-IOS(config-if)#priority-queue cos-map 1 5
CAT6500-IOS(config-if)#
Gi2/6 Gi2/7 Gi2/8

Gi2/6 Gi2/7 Gi2/8
Gi2/6 Gi2/7 Gi2/8
Gi2/6 Gi2/7 Gi2/8
Gi2/6 Gi2/7 Gi2/8
Summary
NOTE
263
Notice that although the number of the priority queue in CatOS will be 3 (or higher), in
Cisco IOS, the priority queue number when using the priority-queue cos-map interface
command is always 1.
It cannot be stressed enough that this chapter is by no means a thorough discussion on

Catalyst QoS; it is merely an overview to provide adequate context for the design chapters
to follow. For a complete discussion of Catalyst QoS, refer to Cisco Catalyst QoS documentation and also the Cisco Press book Cisco Catalyst QoS: Quality of Service in Campus Networks, by Michael Flannagan, Richard Froom, and Kevin Turek.
Summary
For QoS features to be available at line rates in campus environments, they have to be done
in hardware. Porting QoS features to hardware presents many advantages, such as the capability to mark and police at the trafc sources, and the enforcement of trust boundaries and
queuing at (momentarily) congested uplinks and downlinks. However, porting QoS functionality into hardware also presents an undesired caveat from a conguration and management perspective: As hardware varies, QoS also does in functionality and conguration,
leading to signicant complexity in managing campus QoS. To address this caveat, tools
such as MQC and AutoQoS are continuing to be developed to simplify campus QoS.
This chapter examined basic Catalyst QoS models, which, for the most part, can be broken
down into three main areas:
Classication, marking, and mappingThese operations are responsible for

generating the Internal DSCP value by which all subsequent QoS functions will be
referenced. These values may be generated by the congured trust states (untrusted,
trust-cos, trust-ipprec, or trust-dscp), explicit marking policies, or mapping
functions.
Policing and markdownThese functions monitor trafc ows as to whether they

are in prole or out of prole. Markdown actions can be assigned to out-of-prole
trafc (alternatively, out-of-prole trafc can be dropped). Catalyst platforms support
individual, aggregate, or microow (6500 only) policers.
Queuing and droppingHardware queuing models, such as xQyT or xPyQzT

models, were examined from both congestion-management and congestion-avoidance
functionality perspectives.
Having laid a framework for a generic Catalyst QoS model, the model was applied to each
of the main Catalyst platforms (at the time of writing), including the Catalyst 2950, 2970,
3550, 3560, 3750, 4500, and 6500. Conguration syntax and idiosyncrasies were addressed
to lay a basic context for the design discussions to follow.
264
Table 10-5 shows a summary of the QoS features supported by Catalyst platforms.
Table 10-5
Summary of QoS Features Supported by Catalyst Platforms
Platform/
Feature
Catalyst
2950
Catalyst
3550
Catalyst
2970/3560/
3750
Catalyst
4500
Catalyst 6500
QoS on by
default?
Yes
No
No
No
No
Default trust (with

QoS enabled)
Untrusted
Untrusted
Untrusted
Untrusted
Untrusted
Port-based QoS
Yes
Yes
Yes
Yes
Yes
VLAN-based
QoS
No
Yes
No
Yes
Yes
Full DSCP-CoS
maps
No
Yes
Yes
Yes
Yes
DSCP mutation
No
Yes
Yes
No
Yes
Ingress policing
Yes
Yes
Yes
Yes
Yes
Egress policing
No
Yes
Yes
Yes
Yes
Aggregate
policing
No
Yes
Yes
Yes
Yes
Dual-rate policing No
No
No
No
Yes
Microow
policing
No
No
No
No
Yes
Per-user
microow
policing
No
No
No
No
PFC3 only
Policers per port
6 per 10/100
ports; 60 per GE
ports
8 per 10/100 ports;

128 per GE ports
64 per port; 256

per-port ASIC
1020 per
port
1023 per port
Minimum
policing rate/
granularity
1 Mbps
8 kbps
8 kbps
32 kbps
32 kbps
Summary
Table 10-5
265
Summary of QoS Features Supported by Catalyst Platforms (Continued)
Platform/
Feature
Catalyst
2950
Catalyst
3550
Catalyst
2970/3560/
3750
Catalyst
4500
Catalyst 6500
Queuing
structures
4Q1T or 1P3Q1T
10/100 ports:
4Q1T or 1P3Q1T
4Q3T or
1P3Q3T
4Q1T or
1P3Q1T
GE ports: 4Q2T or
1P3Q2T
(Line-card
dependent)
2Q2T
1P2Q1T
1P2Q2T
1P3Q1T
1P3Q8T
1P7Q8T
Priority queue
Q4
Q4
Q1
Q3
(Line-card
dependent)
2Q2T: None
1P2Q1T: Q3
1P2Q2T: Q3
1P3Q1T: Q4
1P3Q8T: Q4
1P7Q8T: Q8
WTD
No
No
Yes
No
No
WRED
No
Yes
No
No
(Line-card
dependent)
2Q2T: No
1P2Q1T: Yes
1P2Q2T: Yes
1P3Q1T: Yes
1P3Q8T: Yes
1P7Q8T: Yes
DBL
No
No
No
Yes
No
AutoQoS
Yes
Yes
Yes
Yes
CatOS: Yes
Cisco IOS: No
266
NOTE
Catalyst 4500 refers to Catalyst 4500 with Sup3-5 running Cisco IOS; Catalyst 6500
refers to Catalyst 6500 with PFC2 (Supervisor 2) or PFC3 (Supervisor 720) running either
CatOS or Cisco IOS. All values were current at the time of writing; check the Cisco
platform documentation for the latest information at http://www.cisco.com/univercd/home/
home.htm.
Further Reading
Books:
Flannagan, Micheal, Richard Froom, and Kevin Turek. Cisco Catalyst QoS: Quality
of Service in Campus Networks. Indianapolis: Cisco Press, 2003.
Cisco Catalyst documentation:
Conguring QoS on the Catalyst 2950 (Cisco IOS Release 12.1[19]EA1):

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat2950/12119ea1/2950scg/
swqos.htm.

http://www.cisco.com/univercd/cc/td/doc/product/lan/c3550/12119ea1/3550scg/
swqos.htm.
Conguring QoS on the Catalyst 3560 (Cisco IOS Release 12.2[20]SE):

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat3560/12220se/
3560scg/swqos.htm.

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat3750/12119ea1/3750scg/
swqos.htm.
Conguring QoS on the Catalyst 4500 (Cisco IOS Release 12.1[20]EW)

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat4000/12_1_20/cong/qos.htm.
Conguring QoS on the Catalyst 6500 (Cisco IOS Release 12.2[17a]SX1):

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat6000/122sx/swcg/qos.htm.
Conguring QoS on the Catalyst 6500 (Cisco CatOS Release 8.2):

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat6000/sw_8_2/
confg_gd/qos.htm.
This chapter discusses wireless LAN QoS tools, including the following:
IEEE 802.11 Distributed Coordination Function (DCF)

IEEE 802.11 Enhanced Distributed Coordination Function (EDCF)
IEEE 802.11 QoS basic service set (QBSS)
IEEE 802.1D classes of service
CHAPTER
11
WLAN QoS Tools

In the past, WLANs mainly were used to transport low-bandwidth data-application trafc.
Today, with the expansion of WLANs into enterprises, small businesses, mobile hotspots,
and home environments, WLANs are used to transport high-bandwidth data applications in
conjunction with time-sensitive multimedia applications. These evolving requirements
have led to the necessity for wireless QoS tools.
However, unique challenges are presented with the use of radio waves as a transmission
medium. For example, radio waves are a shared, half-duplex medium that might not always
be fully controlled (because signals are subject to interference).
Several vendors support proprietary wireless QoS schemes for voice applications. However, a unied approach to wireless QoS would speed up the rate of wireless QoS adoption
and would provide QoS support for time-sensitive applications in multivendor wireless
environments. The IEEE 802.11e working group within the IEEE 802.11 standards committee is dening a wireless QoS standard that was nalized in mid-2004. Other groups,
including the WiFi Alliance, are also drafting standards for wireless QoS.
Cisco Aironet products support QoS based on the IEEE 802.11e draft standard specications as of November 2002. Cisco IOS Software Release 12.2(4)JA for the Cisco Aironet
1100 Series and Cisco Aironet VxWorks Release 12.00T for Cisco Aironet 1200, 350, and
340 Series products support IEEE 802.11e Enhanced Distributed Coordination Function
(EDCF)based wireless QoS.
An example deployment of wireless QoS within a Cisco Aironet environment is shown in
Figure 11-1.
270
Chapter 11: WLAN QoS Tools
Figure 11-1 Cisco Wireless QoS Deployment Example

EDCF-Based
QoS
AP1100
EDCF-Based
QoS
AP1200
Cisco
CallManager
Enterprise
Network
AP provides EDCF-based
mechanisms for downstream
wireless QoS, based upon
handset registration, CoS, or DSCP.
VoIP
Phone
Streaming
Video
QoS for Wireless LANs Versus QoS on Wired LANs

The QoS implementation for wireless LANs differs from QoS implementations on other
(wired) Cisco devices. With QoS enabled, wireless access points (APs) perform the
following QoS functions:
They prioritize packets based on DSCP value, client type (such as a wireless phone),
or the priority value in the 802.1q or 802.1p tag; however, they do not classify packets.
They support mapping by assigning IP DSCP, Precedence, or Protocol values to Layer 2

CoS values; they do not construct internal DSCP values.
They support EDCF (which is discussed in more detail in the section titled IEEE
802.11e EDCF), such as queuing on the (downstream) radio egress port only.
They support only FIFO queuing on the (upstream) Ethernet egress port.
They support Spectralink phones using the class-map IP protocol clause with the
protocol value set to 119.
They support only 802.1Q/P tagged packets. Access points do not support ISL.
They support only the MQC policy-map set cos action.
They prioritize the trafc from voice clients (such as Symbol phones) over trafc from
other clients when the QoS Element for Wireless Phones (which is discussed later in
this chapter in the section titled QoS Basic Service Set Information Element)
feature is enabled.
Upstream Versus Downstream QoS
271
Just as in other media, you might not notice the effects of QoS on a lightly loaded wireless
LAN (WLAN). The benets of QoS become more obvious as the load on the wireless LAN
increases, keeping the latency, jitter, and loss for selected trafc types within an acceptable
range.
QoS on the wireless LAN focuses on downstream prioritization from the access point.
Upstream Versus Downstream QoS

Communication streams in a WLAN environment are over either Ethernet or radio media.
Furthermore, the ows occur in either the upstream or downstream direction, as shown in
Figure 11-2.
Figure 11-2 Ethernet/Radio Upstream/Downstream Flows
Radio Downstream
Ethernet Downstream
Network
Access Point
Radio Upstream
Downstream QoS only
Access Switch
Ethernet Upstream
Bidirectional QoS
Ethernet downstream refers to trafc leaving the switch/router traveling to the access point
(AP). QoS may be applied at this point to prioritize and rate-limit trafc to the AP.
Radio downstream QoS refers to the trafc leaving the AP and traveling to the WLAN
clients. Radio downstream QoS is the primary focus of this chapter.
Radio upstream QoS refers to trafc leaving the WLAN clients and traveling to the AP. No
vendor support is currently available for radio upstream QoS features for WLAN clients.
This support is specied in the 802.11e draft but has not yet been implemented. By providing downstream prioritization from the AP, upstream client trafc is treated as best effort.
A client must compete with other clients for (upstream) transmission and also must compete with best-effort (downstream) transmission from the AP because of the half-duplex
nature of WLAN radio. Under certain load conditions, a client can experience upstream
congestion, and the performance of QoS sensitive applications might be unacceptable,
despite the downstream QoS features on the AP.
Ethernet upstream refers to trafc leaving the AP traveling to the switch. The AP classies
trafc from the AP to the upstream network according to the trafc classication. However,
only FIFO queuing is supported (which is adequate for almost all cases) at this point.
272
IEEE 802.11 DCF

Data frames in 802.11 are sent using the Distributed Coordination Function (DCF). The
DCF is composed of two main components:
Interframe spaces (IFS)

Random backoffs/contention windows (CW)
DCF is used in 802.11 networks to manage access to the RF medium. A baseline

understanding of DCF is necessary to provide context in understanding and deploying
802.11e-based Enhanced Distributed Coordination Function (EDCF).
Interframe Spaces
Interframe Spaces, shown in Figure 11-3, allow 802.11 to control which trafc gets rst
access to the channel when carrier sense declares the channel to be free.
Figure 11-3 IEEE 802.11 DCF Interframe Spaces
DIFS
DIFS
PIFS
Contention Window
SIFS
Busy Medium
Backoff Window
Next Frame
(t)
Slot Time
Defer Access
Select slot and decrement backoff

as long as the medium is idle.
802.11 currently denes three interframe spaces:
Short interframe spaces (SIFS), which are 10 long

Point interframe spaces (PIFS), which are 30 long (SIFS + 1 [20 ] slot time = 30 )
Distributed interframe spaces (DIFS), which are 50 long (SIFS + 2 [20 ] slot
times = 50 )
SIFS
Important frames (such as acknowledgments) await the SIFS before transmitting. There is
no random backoff when using the SIFS because frames using the SIFS are used when
multiple stations would not be trying to send frames at the same time. The SIFS provide a
short and deterministic delay for packets that must go through as soon as possible. SIFS are
not available for use by data frames; only 802.11 management and control frames use SIFS.
IEEE 802.11 DCF
273
PIFS
An optional portion of the 802.11 standard denes priority mechanisms for trafc via PIFS.
No random backoff mechanisms are associated with PIFS because it relies solely on a
polling mechanism to control which station may transmit. The option has not been adopted
widely because of the associated overhead and lack of exibility in its application.
DIFS
Data frames must wait out the DIFS before beginning the random backoff procedure, which
is part of the DCF. This longer wait ensures that trafc using SIFS or PIFS timing always
gets an opportunity to send before any trafc using the DIFS attempts to send. (In other
words, management or priority trafc always gets to send before generic data.)
Random Backoffs/Contention Windows

When a data frame using DCF (shown in Figure 11-4) is ready to be sent, the sending
station goes through the following steps:
1 A random backoff number between 0 and a minimum contention window (CWmin)
is generated.
2 The station waits until the channel is free for a DIFS interval.
3 If the channel is still free, the random backoff number is decremented once for every
slot time (20 ) that the channel remains free.

4 If the channel becomes busy (for example, if another station gets to 0 before your
station), the decrementing of the random backoff number stops and steps 2 through 4
are repeated.
5 If the channel remains free until the random backoff number reaches 0, the frame may
be sent.
Figure 11-4 shows a simplied example of how the DCF process works. (In this simplied
DCF example, no acknowledgments are shown and no fragmentation occurs.)
Figure 11-4 IEEE 802.11 DCF Transmission Example
DIFS
Station A
Station B
Station C
Station D
Station E
DIFS
DIFS
Frame
Def er
Defer
Defer
Frame
Defer
Defer
Frame
Defer
Defer
Backoff Time
Defer
Backoff Time Remaining
Frame
274
The DCF steps illustrated in Figure 11-4 work as follows:

1 Station A successfully sends a frame, and three other stations also want to send frames
but must defer to Station As trafc.

2 When Station A completes transmission, all the stations must still defer to the DIFS.
When the DIFS is complete, stations that want to send a frame can begin decrementing their backoff counters (decrementing by one for every slot time that passes). If
their backoff counters reach 0 and the wire is available, they may send their frame.
3 Station Bs backoff counter reaches 0 before Stations C and D, so Station B begins
transmitting its frame.

4 When Stations C and D detect that Station B is transmitting, they must stop decre-
menting their backoff counters and again defer until the frame is transmitted and a
DIFS has passed.
5 During the time that Station B is transmitting a frame, Station E gets a frame to
transmit, but because Station B is sending a frame, Station E must defer in the same
manner as Stations C and D.
6 When Station B completes transmission and the DIFS has passed, stations with
frames to send begin decrementing their backoff counters again. In this case, Station
Ds backoff counter reaches 0 rst, and the station begins transmission of its frame.
7 The process continues as trafc arrives on different stations.
CWmin, CWmax, and Retries

DCF uses a contention window (CW) to control the size of the random backoff. The
contention window is dened by two parameters:
CWmin
CWmax
The random number used in the random backoff is initially a number between 0 and
CWmin. If the initial random backoff expires without successfully sending the frame, the
station or AP increments the retry counter and doubles the value of the random backoff
window size. This doubling in size continues until the size equals CWmax. The retries
continue until the maximum retries or Time-To-Live (TTL) is reached. This process of
doubling the backoff window often is referred to as a binary exponential backoff; it is
IEEE 802.11e EDCF
275
Figure 11-5 IEEE 802.11 Random Backoff Ranges with Retries
1023 1023 1023
511
aCWmax
255
127
aCWmin
63
31
Retries
IEEE 802.11e EDCF

The current IEEE 802.11e draft describes an Enhanced Distributed Coordination Function
(EDCF). EDCF is an enhancement of DCF, previously described. The main enhancement
is the adjustment of variable CWmin and CWmax random backoff values based upon trafc
classication. This feature is supported in current Cisco Aironet software.
Figure 11-6 shows the principle behind different CWmin values per trafc classication.
All trafc waits the same DIFS, but the CWmin value used to generate the random backoff
number depends upon the trafc classication. High-priority trafc has a smaller CWmin
value, granting it a shorter random backoff value, whereas best-effort trafc has a larger
CWmin value that (on average) generates a longer random backoff value.
276
Figure 11-6 IEEE 802.11e EDCF Random Backoff by Trafc Classication

0
CWmin[10]
CWmin[7]
CWmin[6]
Voice Random Backoff Range

Video Random Backoff Range
Best-Effort Random Backoff Range
DIFS
Contention Window
Busy Medium
Backoff Window
(t)
Next Frame
Slot Time
Defer Access
Decrement backoff as long as the

medium is idle.
Figure 11-7 illustrates how different CWmin values (by trafc class) impact trafc priority
over the WLAN.
Figure 11-7 IEEE 802.11e EDCF Operation Example
DIFS
DIFS
DIFS
DIFS
Frame
Station X
Defer
Voice 1
Defer
Best Effort 1
Voice 2
Defer
Best Effort 2
Defer
Voice 3
Frame
Defer
Defer
Defer
Defer
Defer
Frame
Defer
Defer
Defer
Defer
Frame
Defer
Frame
Backoff Time
Backoff Time Remaining
The process illustrated in Figure 11-7 follows this sequence:

1 While Station X is transmitting its frame, three other stations determine that they
also must send frames. Each station defers (because a frame already was being
transmitted), and each station generates a random backoff number.
IEEE 802.11e EDCF
277
2 Stations Voice 1 and Voice 2 have the trafc classication of voice, so they use an ini-
tial CWmin of 3 and, therefore, generate short random backoff values. Best Effort 1
and Best Effort 2 generate longer random backoff times because their CWmin value
is 31.
3 Voice 1 has the shortest random backoff time and, therefore, starts transmitting rst.
When Voice 1 starts transmitting, all other stations defer. While Voice 1 station is
transmitting, station Voice 3 nds that it, too, needs to send a frame and generates a
random backoff number. However, station Voice 3 defers transmission because of
station Voice 1s transmission.
4 When station Voice 1 nishes transmitting, all stations await the DIFS and then begin
decrementing their random backoff counters again.

5 Station Voice 2 completes decrementing its random backoff counter rst and begins
transmission. All other stations defer.

6 When Station Voice 2 has nished transmitting, all stations wait the DIFS and then
begin decrementing their random backoff counters again.

7 Best Effort 2 completes decrementing its random backoff counter rst and begins
transmission. All other stations defer. This happens even though there is a voice
station waiting to transmit. This shows that best-effort trafc is not starved by voice
trafc because the process of random backoff decrementing eventually brings the
best-effort backoff value down to similar values as initially generated by high-priority
trafc. Additionally, the random process occasionally generates a small random
backoff number for best-effort trafc.
8 When Best Effort 2 nishes transmitting, all stations await the DIFS and then begin
decrementing their random backoff counters again.

9 Station Voice 3 completes decrementing its random backoff counter rst and begins
transmission. All other stations defer.

10 The process continues as other trafc enters the system.
The overall impact of the different CWmin and CWmax values is statistical in nature. It is
sometimes simpler to compare two examples and show the impact of these different values
in the average times that should be generated by the random backoff counters.
If Voice and Best Effort are compared, these trafc categories have default dened CWmin
values of 2 and 5, and CWmax values of 8 and 10, respectively. Each class of trafc can
have a different xed slot time, which species the xed slot backoff interval (0 to 20). The
default values for these parameters for each trafc class are shown in Table 11-1.
From Table 11-1, you can see that that if two stations were contending to transmit voice and
best-effort data at the same time, the station transmitting voice would back off to a random
value between 3 and 4 (Voice CWmin and CWmax), while the station transmitting date
would back off to a value between 5 and 10 (Best Effort CWmin and CWmax). Therefore,
the voice frame would be transmitted before the data frame.
278
Table 11-1
NOTE
Cisco Access Points IEEE 802.11e EDCF CWmin, CWmax, and Fixed Slot Time Default Values
(Cisco IOS 12.2[15]JA)
Class of Service
CWmin
(Min Contention
Window)
CWmax
(Max Contention
Window)
Fixed Slot Time
Best Effort
10
Background
10
Video < 100 ms Latency
Voice < 100 ms Latency
Within EDCF, voice statistically is transmitted before best-effort data. However, voice
might not be serviced ahead of data in every case because of the random element of the
algorithm. Thus, although EDCF does provide a mechanism for preferentially servicing
voice, it is important to note that this is not a strict-priority mechanism.
The average maximum random backoff value gives an indication of how quickly and how
large the random backoff counters can grow in the event of a retransmission. Trafc classes
with the smallest average maximum values behave the most aggressively.
No matter how many times it has retried, voices random backoff delay should not, on
average, be above that of the minimum delay of best-effort trafc. This means that the
average worst-case backoff delay for voice trafc would be the same as the average best
case for best-effort trafc.
NOTE
In this EDCF operation example, all WLAN clients are treated equally for upstream
transmission (from the WLAN clients to the AP).
QoS Basic Service Set Information Element

The WLAN infrastructure devices (such as APs) advertise QoS parameters. WLAN clients
with QoS requirements use these advertised QoS parameters to determine the best AP with
which to associate.
Cisco Aironet software supports the QoS basic service set (QBSS), which is based on IEEE
802.11e draft version 3.3.
Figure 11-8 shows the QBSS information element (IE) advertised by a Cisco AP. The
Channel Utilization eld indicates the portion of available bandwidth currently used to
279
transport data within the WLAN. The Frame Loss Rate eld indicates the portion of
transmitted frames that requires retransmission or is discarded as undeliverable.
Figure 11-8 IEEE 802.11e Draft Version 3.3: QBSS IE Implementation
Element ID
(11)
Length
(6)
Station Count
(2 Octets)
Channel
Utilization
(1 Octet)
Frame
Loss Rate
(1 Octet)
Enabling IEEE 802.11 QBSS adds information to the access point beacons and probe
responses. This information helps some 802.11 phones make intelligent choices about the
access point with which they should associate. Some phones do not associate with an access
point without this additional information. The QBSS can be enabled through the Cisco IOS
CLI (as shown in Example 11-1) or by the web GUI (as shown in Figure 11-11).
Example 11-1 Enabling IEEE 802.11E QBSS Through Cisco IOS CLI on a Cisco Aironet 1100 AP
AP1100(config)#dot11
dot11 phone

IEEE 802.11e classes of service borrow the classes originally dened in IEEE 802.1D, as
illustrated in Table 11-2.
Table 11-2

user_priority
Abbreviation
Traffic Type
BK
Background
Spare
0 (default)
BE
Best effort
EE
Excellent effort
CL
Controlled load
VI
Video, < 100 ms latency and jitter
VO
Voice, < 10 ms latency and jitter
NC
Network control
Compatibility issues are presented with the use of these 802.1D classes of service with
respect to Cisco defaults.
For example, Cisco telephony devices, such as IP phones, follow the IETF recommendations (RFC 3268) of marking real-time trafc (voice) to DSCP EF. The simplest mapping
280
from (6-bit) DSCP values to (3-bit) IP Precedence, CoS, or even MPLS EXP values is to
use the three most signicant bits of the DSCP for the IP Precedence/CoS/MPLS EXP values.
This would map DSCP EF to CoS 5, which is what Cisco devices do. However, according
to the (aging) IEEE 802.1D standard, voice should be mapped to CoS 6. To compensate for
this seeming inconsistency between standards, Cisco Aironet software (by default when
QoS is enabled) maps IEEE 802.1Q/p CoS 5 to IEEE 802.11e CoS 6 so that voice trafc is
given highest priority over the WLAN.
QoS Operation on Cisco APs

When you enable QoS, the access point queues packets based on the Layer 2 class of
service value for each packet. The access point applies QoS policies in this order:
NOTE
Packets already classied. When the access point receives packets from a QoSenabled switch or router that already has classied the packets with nonzero 802.1Q/P
user_priority values, the access point uses that classication and does not apply other
QoS policy rules to the packets. An existing classication takes precedence over all
other policies on the access point.
Even if you have not congured a QoS policy, the access point always honors tagged
802.1P packets that it receives over the radio interface.
QoS Element for Wireless Phones setting. If you enable the QoS Element for
Wireless Phones setting, trafc from voice clients takes priority over other trafc,
regardless of other policy settings. The QoS Element for Wireless Phones setting takes
precedence over other policies, second only to previously assigned packet
classications.
Policies you create on the access point. QoS policies that you create and apply to
VLANs or to the access point interfaces are third in precedence after previously
classied packets and the QoS Element for Wireless Phones setting.
Default classication for all packets on VLAN. If you set a default classication for
all packets on a VLAN, that policy is fourth in the precedence list.
Conguring QoS on Cisco APs
281

QoS is disabled by default; however, the radio interface always honors tagged 802.1P
packets even when you havent congured a QoS policy. QoS can be congured through
either the Cisco IOS (MQC) CLI or the web-interface conguration utility provided by the
AP software.
The Cisco IOS MQC CLI provides a very familiar syntax for QoS conguration on Cisco
APs. In Example 11-2, DSCP EF is mapped to IEEE 802.1D CoS 6 (the IEEE denition
for voice CoS) and is applied to the IEEE 802.11B radio interface (because this mapping
already is enabled by default, this part of the policy is actually moot and is provided for
syntax example purposes only).
Furthermore, the CWmin and CWmax values for the Best Effort and Background classes
have been tuned (away from the defaults shown in Table 11-1) to assign Background trafc
a signicantly less-than best-effort service (compatible to the Scavenger service).
Specically, Best-Effort (Trafc Class 0) is set with a CWmin of 5 and a CWmax of 8, and
Background trafc (Trafc Class 1) is set to a CWmin of 9 and a CWmax of 10.
Example 11-2 Conguring QoS on a Cisco Aironet 1100 AP
AP1100(config)#class-map match-all VOICE
AP1100(config-cmap)# match ip dscp ef
! Matches voice by DSCP EF
AP1100(config-cmap)#policy-map AP-DOWNSTREAM
AP1100(config-pmap)# class VOICE
AP1100(config-pmap-c)#
set cos 6
! Sets IEEE 802.11e CoS for voice
AP1100(config-pmap-c)#interface Dot11Radio0
AP1100(config-if)# traffic-class best-effort cw-min 5 cw-max 8 fixed-slot 2
AP1100(config-if)# traffic-class background cw-min 9 cw-max 10 fixed-slot 6
AP1100(config-if)# service-policy output AP-DOWNSTREAM
Alternatively, QoS for APs can be congured from the web GUI. From the home page,
choose Services > QoS to bring up the main QoS conguration screen, shown in Figure 11-9.
282
Figure 11-9
Cisco Aironet Software Web-Based Congurator: QoS Main Screen
CWmin and CWmax values for each of the four queues can be adjusted from the 802.11B
Access Categories tab (see Figure 11-10).
Advanced options, such as enabling the QBSS information element and disabling the
default AVVID priority mapping (of 802.1Q/p CoS 5 to IEEE 802.11e CoS 6), can be
selected from the Advanced screen, shown in Figure 11-11.
283
Figure 11-10 Cisco Aironet Software Web-Based Congurator: 802.11B Access Categories Screen
NOTE
The option for mapping IEEE 802.1Q/p Ethernet packets marked with CoS 5 to IEEE
802.11e CoS 6 (shown in the Advanced screen) is enabled by default. This enables simpler
integration with existing AVVID networks already using 802.1Q/p CoS 5 to identify voice
trafc. This option also can be disabled (if necessary) through the Cisco IOS CLI with the
no dot11 priority-map avvid command.
284
Figure 11-11 Cisco Aironet Software Web-Based Congurator: QoS Advanced Screen
Summary
As WLANs continue to proliferate into enterprises, into small/medium businesses, into
mobile hotspots, and even within home networks, their use is expanding rapidly to carry
multiservice trafc, including time-sensitive/interactive applications such as wireless VoIP
and video.
To ensure that the WLAN does not become the weakest link in an end-to-end QoS chain,
standards bodies such as the IEEE 802.11e working group are formalizing specications to
deliver granular service levels across WLAN media.
This chapter overviewed IEEE 802.11 DCF operation, including the function of interframe
spaces and contention windows. Building on this review, IEEE 802.11e EDCF was introduced; it allows for contention window variance for different trafc classes. EDCF operation was examined to demonstrate how contention window variance translates into relative
trafc class priorities on the WLAN. The IEEE 802.1e QBSS information element also was
covered in brief.
Further Reading
285
Cisco AP software QoS operation and conguration was introduced, including the Cisco
IOS MQC CLI for AP QoS and the web-based GUI screens required to set wireless accesspoint QoS.
Further Reading
IEEE 802.11 working group: http://www.ieee802.org/11/.
IEEE 802.11e contention-based channel access (EDCF) performance evaluation:

http://path.berkeley.edu/dsrc/reading/03-ICC-EDCF.pdf.
Conguring QoS for WLANs: Cisco IOS Software 12.2.15JA documentation: http://
www.cisco.com/univercd/cc/td/doc/product/wireless/airo1100/accsspts/i12215ja/
i12215sc/s15qos.htm.
IEEE 802.11e status: http://grouper.ieee.org/groups/802/11/Reports/tge_update.htm.

IEEE 802.11e WLAN QoS: http://www.eecs.umich.edu/~shchoi/
My_Papers_Published/Conferences/02-EW.pdf.
PA R T
III
LAN QoS Design

Part III of this book provides an in-depth discussion of QoS considerations and designs for the access,
distribution, and core layers of an enterprise campus network. Detailed QoS designs for the Cisco Catalyst
2950, 2970, 3550, 3560, 3750, 4500, and 6500 switch platforms are presented.
The chapter in this part of the book is as follows:
Chapter 12
Campus QoS Design
This chapter discusses QoS considerations and design principles for the access, distribution, and core layers of an enterprise campus network. These include access edge models
for trusted, untrusted, and conditionally trusted endpoints, along with queuing models that
vary on a per-platform or per-line card basis.
Detailed QoS designs for the following Catalyst switch platforms are presented:
Catalyst 2950
Catalyst 2970
Catalyst 3550
Catalyst 3560
Catalyst 3750
Catalyst 4500 (Supervisors II+ through V)
Catalyst 6500 (Supervisor 2/PFC2 and Supervisor 720/PFC3)
CHAPTER
12
Campus QoS Design

The case for QoS in WANs and VPNs is, for the most part, self evident simply because of
the low-bandwidth links involved compared to the high-bandwidth requirements of most
applications. However, in Gigabit and Ten Gigabit campus LAN environments, bandwidth
is so plentiful that sometimes the need for QoS is overlooked or outright challenged.
This is often the case when network administrators equate QoS with queuing only. But,
as has been shown, the QoS toolset extends considerably beyond just queuing tools. In
addition to queuing, classication, marking, and policing are all important QoS functions
that are performed optimally within the campus network, particularly at the access-layer
ingress edge (access edge).
Three important QoS design principles come into play to make the case for deploying
campus QoS policies.
The rst is that applications should be classied and marked as close to their sources
as technically and administratively feasible. This principle promotes end-to-end
Differentiated Services and Per-Hop Behaviors.
Sometimes endpoints can be trusted to set CoS and DSCP markings correctly, but, in most,
cases it is not a good idea to trust markings that users can set on their PCs (or other similar
devices). This is because users easily could abuse provisioned QoS policies if permitted to
mark their own trafc. For example, if DSCP EF received priority services throughout the
enterprise, a user could congure his PC to mark all his trafc to DSCP EF right on the
NIC, thus hijacking network priority queues to service non-real-time trafc. Such abuse
could ruin the service quality of Real-Time applications (such as VoIP) throughout the
enterprise. For this reason, the clause as close as . . . administratively feasible is included
in the design principle.
290
Chapter 12: Campus QoS Design
A second important QoS design principle relating to access-edge QoS design is that
unwanted trafc ows should be policed as close to their sources as possible. There is little
sense in forwarding unwanted trafc only to police and drop it at a subsequent node. This
is especially the case when the unwanted trafc is the result of DoS or worm attacks. The
overwhelming volumes of trafc that such attacks create can readily drive network device
processors to their maximum levels, causing network outages.
The third important design principle is that QoS always should be performed in hardware
rather than software when a choice exists. Cisco IOS routers perform QoS in software,
which places additional taxes on the CPU (depending on the complexity and functionality
of the policy). Cisco Catalyst switches, on the other hand, perform QoS in dedicated
hardware ASICS and, as such, do not tax their main CPUs to administer QoS policies.
Therefore, complex QoS policies can be applied at Gigabit and Ten Gigabit Ethernet line
speeds in these switches.
For these reasons, QoS policies, such as classication and marking policies to establish and
enforce trust boundaries, as well as policers to protect against undesired ows, should be
enabled at the access edge of the LAN. However, the need for queuing in the campus should
not be dismissed.
Some studies have shown that 95 percent of the time, campus access-layer links are utilized
at less than 5 percent of their capacity. Such underutilization permits campus networks to
be designed to accommodate oversubscription among access, distribution, and core layers.
Oversubscription allows for uplinks to be utilized more efciently and, more important,
reduces the overall cost to build the campus network. Some typical values for campus
oversubscription are 20:1 for the access-to-distribution layers and 4:1 for the distributionto-core layers, as shown in Figure 12-1.
Figure 12-1 Typical Campus Oversubscription Ratios
Core
Si
Si
Typical 4:1
Oversubscription
Distribution
Si
Si
Typical 20:1
Oversubscription
Access
291
Under normal operating conditions, it is quite rare for campus networks to experience
congestion and, thus, have a need for queuing. When congestion does occur, it is usually
momentary and not sustained, as at a WAN edge.
Nonetheless, critical applications, such as VoIP, require service guarantees regardless of
network conditions. The only way to provide service guarantees is to enable queuing at any
node that has the potential for congestionregardless of how rarely this actually might
occur. Because of the oversubscription ratios just discussed, the potential for congestion
exists in campus uplinks. Furthermore, the potential for congestion also exists in campus
downlinks because of mismatches (such as Gigabit EthernettoFast Ethernet links). In
both cases, the only way to ensure service guarantees in the event of congestion is to enable
queuing at these points.
So far, this discussion for enabling queuing within the campus has revolved around network
requirements under normal operating conditions. However, probably the strongest case for
enabling QoS within the campus is to consider what happens under abnormal network
conditions, such as a DoS or worm attack. During such conditions, network trafc increases
exponentially until links are utilized fully. Without QoS, applications are drowned out by
the worm-generated trafc, causing DoS through unavailability. On the other hand, when
QoS policies are enabled within the campus (as detailed later in this chapter), VoIP, critical
applications, and even Best-Effort trafc is protected and serviced if a worm attack occurs,
thus maintaining the networks availability.
In such worst-case scenarios, the intrinsic interdependencies of network QoS, highavailability, and security are clearly manifest.
So where is QoS required in the campus?
Access switches require the following QoS policies:
Appropriate (endpoint-dependent) trust policies

Classication and marking policies
Policing and markdown policies
Queuing policies
Distribution and core switches require the following:
DSCP-trust policies
Queuing policies
Optionally, per-user microow policing policies (on distribution-layer Catalyst 6500s
with Supervisor 720s only)
292
Figure 12-2 summarizes these recommendations.

Figure 12-2 Where QoS Is Required Within the Campus
Core
Si
Si
Si
Si
Distribution
Access
Access Edges: Trust, Classification, Marking, Policing, and Queuing Policies

Interswitch Links: DSCP-Trust and Queuing Policies
Optional (C6500-PFC3 Only): Per-User Microflow Policing on Uplinks from Access Layer
Some important considerations to keep in mind when dening campus QoS designs follow:
DoS/worm-mitigation strategies
Call-signaling TCP/UDP ports in use
Access-edge trust models
WAN aggregator and branch router connections
Each of these concerns is discussed in the following sections.
DoS/Worm-Mitigation Strategies
A proactive approach to mitigating DoS/worm ooding attacks within campus environments is to respond immediately to out-of-prole network behavior that indicates a DoS or
worm attack using access-layer policers. Such policers could meter trafc rates received
from endpoint devices, and when these exceed specied watermarks (at which point they
no longer are considered normal ows), these policers could mark down excess trafc.
DoS/Worm-Mitigation Strategies
293
In this respect, the policers would be fairly dumb. They would not be matching specic
network characteristics of specic types of attacks, but they simply would be metering
trafc volumes and responding to abnormally high volumes as close to the source as
possible. The simplicity of this approach negates the need for the policers to be programmed
with knowledge of the specic details of how the attack is being generated or propagated.
It is precisely this dumbness of such access-layer policers that enables them to maintain
relevancy as worms mutate and become more complex: The policers dont care how the
trafc was generated, what it looks like, or what port it is being served onall they care
about is how much trafc is being put onto the wire. Therefore, they continue to police even
advanced worms that continually change the tactics of how trafc is being generated.
For example, in most enterprises, it is abnormal (within a 95-percent statistical condence
interval) for PCs to generate sustained trafc in excess of 5 percent of their links capacity.
In the case of a Fast Ethernet switch port, this would mean that it would be unusual in most
organizations for an end users PC to generate more than 5 Mbps of uplink trafc on a
sustained basis.
NOTE
It is important to recognize that this value ( 5 percent) for normal access-edge utilization
by endpoints is just an example value. This value likely varies from industry vertical to
vertical, and from enterprise to enterprise. To keep things simple, this 5-percent value is
being used in the examples presented in this design chapter.
It is important to recognize that what is being proposed is not policing all trafc to 5 Mbps
and automatically dropping the excess. If that was the case, there would not be much reason
to deploy Fast Ethernet or Gigabit Ethernet switch ports to endpoint devices because even
10BASE-T Ethernet switch ports would have more uplink capacity than a 5 Mbps policerenforced limit. Furthermore, such an approach supremely would penalize legitimate trafc
that did exceed 5 Mbps on an FE switch port.
A less draconian approach is to couple access-layer policers with hardware and software
(campus, WAN, VPN) queuing polices, with both sets of policies provisioning for a lessthan best-effort Scavenger class.
This could work by having access-layer policers mark down out-of-prole trafc to DSCP
CS1 (Scavenger) and then have all congestion-management policies (whether in Catalyst
hardware or in Cisco IOS Software) provision a less-than best-effort service for any trafc
marked to CS1 during periods of congestion.
294
Scavenger-Class QoS Operation

This section examines how the Scavenger-class QoS strategy for DoS/worm mitigation
might work, both for legitimate trafc that exceeds the access-layer policers watermark
and also in the case of illegitimate excess trafc (the result of a DoS or worm attack).
In the former case, imagine that the PC generates more than 5 Mbps of trafcperhaps
because of a large le transfer or backup. Within the campus, there is generally abundant
capacity to carry the trafc, so congestion (under normal operating conditions) is rarely, if
ever, experienced. This is usually the case because the uplinks to the distribution and core
layers of the campus network are typically Gigabit Ethernet and would require (at least)
1000 Mbps of trafc from the access-layer switch to experience congestion.
If the trafc were destined for the far side of a WAN or VPN link (these are rarely greater
than 5 Mbps in speed), dropping would occur even without the access-layer policer simply
because of bottlenecks resulting from the campus/WAN speed mismatch. TCPs sliding
windows mechanism eventually would nd an optimal speed (less than 5 Mbps) for the le
transfer.
To summarize, access-layer policers that mark down out-of-prole trafc to Scavenger
(CS1) would not affect legitimate trafc, aside from the obvious re-marking. No reordering
or dropping would occur on such ows as a result of these policers (that would not have
occurred anyway).
In the latter case, the effect of access-layer policers on trafc caused by DoS or worm
attacks is quite different. As hosts become infected and trafc volumes multiply, congestion
might be experienced even within the campus. If just 11 end-user PCs on a single switch
begin spawning worm ows to their maximum Fast Ethernet link capacities, a GE uplink
from the access-layer switch to the distribution-layer switch will become congested, and
queuing and reordering will engage. At such a point, VoIP and critical data applications
and even Best-Effort applicationswould gain priority over worm-generated trafc
(because this Scavenger-marked trafc would be dropped the most aggressively); network
devices would remain accessible for the administration of patches, plugs, and ACLs
required to fully neutralize the specic attack.
295
WAN links also would be protected: VoIP, Critical Data, and even Best-Effort ows
would continue to receive priority over any trafc marked down to Scavenger/CS1. This
is a huge advantage because generally WAN links are the rst to be overwhelmed by DoS/
worm attacks.
The bottom line is that access-layer policers signicantly will mitigate network trafc
generated by DoS or worm attacks.
It is important to recognize the distinction between mitigating an attack and preventing it
entirely: The strategy being presented will not guarantee that no DoS or worm attacks ever
will happen; it only reduces the risk and impact that such attacks could have on the campus
network infrastructure and then, by extension, the WAN and VPN network infrastructure.

In this design chapter, to keep the examples relatively simple, only Skinny Call Control
Protocol (SCCP) ports (TCP ports 20002002) are used to identify call-signaling protocols.
However, SCCP is by no means the only call-signaling protocol used in IP Telephony
environments. Table 12-1 shows many of the TCP/UDP ports used in a Cisco CallManager
environment.
All relevant call-signaling ports that are required for a given IPT environment are recommended to be included in the access lists that identify call-signaling protocols. Furthermore, rewalls protecting CallManagers should enable additional ports to provide the
supplementary services that CallManagers provide or require.
Remote
Source
Port
CallManager
Destination
Port
DTC
CallManager Remote Device

Source
Destination
Port
Port
TCP 135
Remote Devices
CallManagers in the
same cluster
SSH
TCP 22
Secure Shell client
Telnet
TCP 23
Telnet client
DNS
UDP 53
DNS servers
UDP 67
DHCP server
DHCP
UDP 68
DHCP
UDP 68
UDP 67
DHCP client
TFTP
UDP 69
HTTP
TCP 80
Administrator/user
web browsers
OSI (DAP,
DSP, DISP)
TCP or UDP 120
DCD Directory
NTP
UDP 123
WINS
UDP 137 to 139
SNMP
UDP 161
SNMP Trap
Notes
Dynamic ports used

after initial connect
CCMAdmin and
CCMUser pages
WINS Server
Windows Internet
Name Service
TCP 389
Directory Services
When integrated
with Corporate
Directory
UDP 162
LDAP
TCP 389
HTTPS/SSL
TCP 443
SMB
TCP 445
TCP 445
CallManagers in the
same cluster
Syslog
TCP 514
UDP 514
Syslog service
Protocol
Example TDP/UDP Ports Used in a Cisco CallManager Environment
296
Table 12-1
Table 12-1
Example TDP/UDP Ports Used in a Cisco CallManager Environment (Continued)
Protocol
Remote
Source
Port
CallManager
Destination
Port
RMI
TCP 1099 to
1129
MS SQL
TCP 1433

Source
Destination
Port
Port
TCP 1024 to
4999
TCP 1720
H.323, H.225/ICT
TCP 1024 to
4999
H.323, H.245
TCP 1024 to
4999
TCP 1433
CallManagers in the
same cluster
TCP 1719
Gatekeeper RAS
CallManager earlier
than 3.3, Cisco
Conference
Connection
TCP 1719
Gatekeeper RAS
CallManager 3.3
TCP 1720
H.323 gateways,
anonymous device
Cisco Conference
Connection,
non-Gatekeepercontrolled H.323
trunk
CallManager
Gatekeepercontrolled H.323
trunks
TCP 1024 to
4999
CallManager 3.3
CallManager H.323
gateways,
anonymous device,
H.323 trunks
continues
H.323, H.225
Notes
RMI Service
Attendant Console
H.323 RAS
H.323 RAS
Remote Devices
297

CallManager
Destination
Port

Source
Destination
Port
Port
Remote Devices
H.323, H.245
TCP 11000 to
11999
Cisco IOS H.323

gateways, Cisco
Conference
Connection
SCCP
TCP 2000
Skinny Clients (IP

phones)
Skinny Gateway
(Analog)
TCP 2001
Analog Skinny
Gateway
Skinny Gateway
(Digital)
TCP 2002
Digital Skinny
Gateway
MGCP Control
UDP 2427
MGCP Gateway
Control
MGCP Backhaul
TCP 2428
MGCP Gateways
Backhaul
RTS Serv
2500
Cisco Extended
Service
TCP 2551
Active/backup
determination
Cisco Extended
Service
TCP 2552
DB change
notication
RIS Data
Collector
TCP 2555
Inter-RIS
communication
RIS Data
Collector
TCP 2556
Used by clients (IIS)

to communicate with
RIS
Notes
Protocol
Remote
Source
Port
298
Table 12-1
Table 12-1
Protocol
Remote
Source
Port
CallManager
Destination
Port

Source
Destination
Port
Port
Remote Devices
Notes
Connects with CTI
Manager; used by
IVR, CCC, PA,
Cisco SoftPhone,
CRS, ICD, IPCC,
IPMA, Attendant
Console, and any
other application that
utilizes the TAPI or
JTAPI plug-in; TSP
TCP 2748
TAPI/JTAPI
applications
IPMA Service
TCP 2912
IPMA Assistant
Console
Media Streaming
Application
UDP 3001
Change notication
SCCP
TCP 3224
Media resources
MS Terminal
Services
TCP 3389
Windows Terminal
Services
Entercept HID
Agent
CallManager SIP
TCP/UDP 5060
VNC HTTP
Helper
TCP 580x
TCP 5000
Host Intrusion
Detection Console
TCP 5060
SIP Trunk default

port
Conference bridges,
Xcoders
Can use TCP 1024 to

65,535
Remote control
continues
CTI/QBE
299

CallManager
Destination
Port

Source
Destination
Port
Port
Remote Devices
Notes
VNC Display
TCP 690x
Virtual Network
Computer Display
Remote control
CallManager
Change
Notication
TCP 7727
CallManager change
notication, Cisco
database layer
monitor, Cisco
TFTP, Cisco IP
media streaming,
Cisco TCD, Cisco
MOH
Real-time change
notication
IPMA Service
TCP 8001
IP Manager
Assistant
Change notication
ICCS
TCP 8002
CallManagers in the
same cluster
Intracluster
communication
CTIM
TCP 8003
Cisco Tomcat
TCP 8007
Web requests
Cisco Tomcat
TCP 8009
Web requests
Cisco Tomcat
TCP 8111
IIS, web requests to

IPMA worker thread
Cisco Tomcat
TCP 8222

EM application
worker thread
Cisco Tomcat
TCP 8333

WebDialer
application worker
thread
TCP 8002
Protocol
Remote
Source
Port
300
Table 12-1
Table 12-1
Protocol

Remote
Source
Port
CallManager
Destination
Port

Source
Destination
Port
Port
Remote Devices
Notes
Used for Directory
services, application
authentication/
conguration,
SoftPhone directory,
user directory
TCP 8404
Embedded Directory
Services
Cisco Tomcat
TCP 8444

EM service worker
thread
Cisco Tomcat
TCP 8555

Apache SOAP
worker thread
Cisco Tomcat
TCP 8998
Web requests
Cisco Tomcat
TCP 9007

CAR worker thread
RTP
Cisco SNMP
Trap Agent
UDP 16384
to 32767
UDP 16384 to 32767 Voice media
UDP 61441
Cisco Alarm
Interface
IP IVR media, CCC

IVR media, Cisco
SoftPhone, Media
Streaming
Application
Receives some
SNMP alarm in
XML format
DC Directory
301
302
Access-Edge Trust Models

A primary function of access-edge policies is to establish and enforce trust boundaries. A
trust boundary is the point within the network where markings (such as CoS or DSCP)
begin to be accepted. Previously set markings are overridden (as required) at the trust
boundary.
The design objective relating to trust boundaries is to enforce these as close to the endpoints
as technically and administratively possible. This is illustrated in Figure 12-3.
Figure 12-3 Establishing Trust Boundaries
Endpoints
Access
Distribution
Core
Si
Si
Si
Si
WAN Aggregators
2
3
Trust Boundary
1 Optimal Trust Boundary:
Trusted Endpoint
2 Optimal Trust Boundary:

Untrusted Endpoint
3 Suboptimal Trust Boundary
The denition of the trust boundary depends on the capabilities of the endpoints that are
being connected to the access edge of the LAN. Three main categories of endpoints exist,
as they relate to trust boundaries:
Trusted endpoints
Untrusted endpoints
Conditionally trusted endpoints
Each of these categories of endpoints is discussed in detail in the following sections.
Trusted Endpoint Models

Trusted endpoints have the capabilities and intelligence to mark application trafc to the
appropriate CoS and DSCP values. Trusted endpoints also possess the capability to re-mark
trafc that previously was marked by an untrusted device. For the most part, trusted
endpoints are not mobile devicesthe switch port that they are plugged into generally does
not change.
NOTE
303
IP phones, which often change switch ports as users move, more appropriately are included
in the category of conditionally trusted endpoints.
Examples of trusted endpoints include these:
Analog gatewaysThese devices are deployed to connect analog devices (such as

fax machines, modems, TDD/TTYs, and analog phones) to the VoIP network so that
the analog signals can be packetized and transmitted over the IP network. Examples
of analog gateways include these:
The NM-1V and NM2-V network modules, which support either
High-Density or Low-Density Voice/Fax Interface Cards or VICs
The Cisco Communication Media Module (CMM) line card
The Catalyst 6500 Analog Interface Module (WS-X6624-FXS)
The Cisco VG224 and VG248 Cisco IOSbased voice gateways
The Cisco ATA 186/188 Analog Telephony Adaptors
IP conferencing stationsThese devices are designed for meeting-room VoIP

conferencing. Essentially, they are specialized IP phones with 360 microphones and
advanced speakerphones. Examples of such devices include the Cisco 7935 and 7936.
Videoconferencing gateways and systemsThese devices transmit interactive

video across the IP network. Examples of such devices capable of setting DSCP
markings include the Cisco IP/VC 3511, 3521, 3526, and 3540 videoconferencing
gateways and systems.
Video surveillance unitsThese (third-party) devices are used for security and
remote-monitoring purposes over an IP (as opposed to a closed-circuit) network.
These might support DSCP marking, in which case they can be considered trusted
endpoints.
ServersCertain servers, within the data center or otherwise, might be capable of

correctly marking their trafc on their NICs. In such cases, the network administrator
can choose to trust such markings. However, enforcing such a trust boundary requires
cooperation between network administrators and system or server administrators, an
alliance that is often fragile, at best, and usually involves considerable nger pointing.
Additionally, network administrators should bear in mind that the majority of DoS/
worm attacks target servers. Infected servers not only might spew profuse amounts of
trafc onto the network, but, in such cases, they might do so with trusted markings.
Theres no hard-and-fast rule that will apply to every situation. Some administrators
prefer to trust certain servers, like Cisco CallManagers, due to the large number of
ports that may be in use to provide services (refer to Table 12-1) rather than administer
complex access lists. In either case, consider the tradeoffs involved when deciding
whether or not to trust a server.
304
Wireless access pointsSome wireless APs have the capability to mark or re-mark
802.1p CoS or DSCP values and, therefore, qualify as trusted endpoints. Examples
include Cisco Aironet 350, 1100, and 1200 series APs.
Wireless IP phonesMobile wireless IP phones can mark DSCP values for VoIP and
Call Signaling and pass these on to the wireless AP they are associated with.
Examples include the Cisco 7920G wireless IP phone.
When trusted endpoints are connected to a switch port, typically all that is required is
enabling the following interface command: mls qos trust dscp.
Optionally, if the trafc rate of the trusted application is known, the network administrator
could apply an access-layer policer to protect against out-of-prole rates, in case the trusted
endpoint somehow is compromised. For example, consider the case of an IP videoconferencing station that transmits 384 kbps of video (not including Layers 2 through 4 overhead)
and correctly marks this trafc to DSCP AF41. An access-edge ingress policer could be
applied to the switch port that this IP/VC station is connected to and could be congured to
trust up to 500 kbps (allowing for Layers 2 through 4 overhead and policer granularity) of
Interactive-Video trafc (marked AF41); excess trafc could be marked to CS1. Such a policy
would prevent network abuse if another device was inserted into the path (perhaps through
a hub) or if the trusted endpoint itself became compromised.
Untrusted Endpoint Models

As previously mentioned, trusting end users and their PCs is generally a bad idea because
newer operating systems such as Windows XP and Linux make it relatively easy to set CoS
or DSCP markings on PC NICs. Such markings can be set deliberately or even inadvertently.
In either case, improperly set QoS markings could affect the service levels of multiple users
within the enterprise and make troubleshooting a nightmare. Also, marking application trafc on server NICs has disadvantages (discussed in the previous section) that might make it
preferable to treat these as untrusted devices.
Although client PCs and data center servers are related and complimentary, they also have
unique considerations that affect their classication and marking policies, so they are
examined individually next.
Untrusted PC with SoftPhone Model

It generally is recommended not to trust end-user PC trafc. However, some PCs might
be running applications that critically require QoS treatment. A classic example is a PC
running Cisco IP SoftPhone. In such a case, the critical application would need to be
identied through access lists and marked or re-marked at the access edge. Re-marking can
be done with either the MLS QoS set ip dscp command or with a policer.
NOTE
305
In this context, SoftPhone can be used to refer to any PC-based IP telephony application.
A policer is recommended in this case because limits on the amount of trafc being marked
could be imposed (again, to prevent abuse). SoftPhones can use regular G.711 codecs (in
which case, 128 kbps is adequate), or they can be congured to use a G.722 (wide codec,
in which case 320 kbps is required). The tighter the policer is, the better (provided that
adequate bandwidth has been allocated for the applications requirements).
Additionally, the UDP ports used by SoftPhone can be dened explicitly within the application (instead of simply picking random ports within the UDP range of 16,383 to 32,767).
This is recommended, because this would allow for a more granular access list to match
legitimate SoftPhone trafc, thereby tightening the overall security of the policy.
Figure 12-4 illustrates the logic of such an access-edge policer marking SoftPhone trafc
from an untrusted PC endpoint.
Figure 12-4 Untrusted Endpoint Policing: PC with SoftPhone Example
Start
UDP
16384 to
32767
Yes
No
No
TCP
20002002
No
Yes
Re-Mark to DSCP CS1

and Transmit
Yes
Yes
No
Re-Mark to DSCP CS3

and Transmit
Re-Mark to DSCP CS1
and Transmit
Yes
No
Re-Mark to DSCP EF
and Transmit
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
The syntax for implementing such a policer might vary slightly from platform to platform,
as is detailed in the coming platform-specic sections.
Untrusted Server Model

As with PCs, servers are subject to attack and infection by worms and viruses and also
should be policed for the amounts of trafc they admit onto the network. Granted, the
values are much greater than with PC endpoints, and it is up to network administrators to
prole trafc patterns from servers to baseline what is normal and abnormal behavior.
306
As an example, assume that a single server is running multiple applicationsin this case,
SAP (TCP ports 3200 to 3203 and also 3600), Lotus Notes (TCP port 1352), and IMAP
(TCP ports 143 and 220). SAP is considered a Mission-Critical application and (until Call
Signaling marking on IP Telephony equipment fully migrates from DSCP AF31 to CS3)
should be marked to DSCP 25. Lotus Notes is classed as a Transactional Data application
and should be marked to DSCP AF21; IMAP is considered a Bulk Data application and
should be marked to DSCP AF11.
Application baselining has shown that, 95 percent of the time, the trafc rates for SAP,
Lotus Notes, and IMAP are less than 15 Mbps, 35 Mbps, and 50 Mbps, respectively. No
other trafc should emanate from the server; to ensure this, a nal policer to catch any other
type trafc is included. Remember, if legitimate trafc temporarily exceeds these values,
no dropping or reordering of packets will occur. However, if this server becomes infected
and begins sending sustained trafc in excess of these normal rates, the excess would be
subject to aggressive dropping in the event of link congestion. Figure 12-5 shows the logic
of such a policer.
Figure 12-5 Untrusted Endpoint Policing: Multiapplication Server Example
Start
MCD ACLs
Mission-Critical
Data ACLs
Bulk
Data ACLs
Yes
No
Yes
35 Mbps
Yes
Re-Mark to DSCP AF21

and Transmit
Re-Mark to DSCP CS1
and Transmit
Yes
Yes
Re-Mark to DSCP 25
and Transmit
Re-Mark to DSCP CS1
and Transmit
No
No
Bulk Data
ACLs
15 Mbps
No
No
TD ACLs
Transactional
Data ACLs
Yes

and Transmit
Re-Mark to DSCP CS1
and Transmit
No
Yes
No
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
One of the critical concerns in deploying QoS designs for untrusted servers is to remember
that the applications usually are identied by source ports, not destination ports (as is the
case with client-to-server access lists).
Thus, the access lists for server-to-client trafc becomes this:
permit [ tcp | udp ] any [ eq | range ] any
307
instead of the access list for client-to-server trafc:

permit [ tcp | udp ] any any [ eq | range ]
This is a subtle but critical difference.
Conditionally Trusted Endpoint(s) Models

One of the main business advantages of IP Telephony is the simplicity and related cost
savings of user adds, moves, and changes. To move, all a user has to do is pick up the IP
phone; plug it into the new location. To move, all a user has to do is pick up the IP phone,
plug it into the new location, and carry on business as usual. If the infrastructure supports
inline power, it is literally a matter of unplugging a single RJ-45 cable and plugging it in at
the new location.
IP phones are trusted devices; PCs are not. This presents a problem when it comes to provisioning trust in a mobile environment. Consider the following example: Port A is congured to trust the endpoint connected to it, which initially is an IP phone. Port B is congured
not to trust the endpoint connected to it, which initially is a PC. As the result of a move,
these endpoints end up plugged into the opposite ports. This breaks the VoIP quality of calls
made from the IP phone (now plugged into untrusted Port B) and opens the network for
unintentional or deliberate abuse of provisioned QoS by the PC (now plugged into the trusted
Port A).
One solution is to place a call to the networking help desk when the move is scheduled
so that the switch ports can be recongured to trust or untrust the endpoints, as required.
However, this approach would dampen the mobility business advantage of IP Telephony
because manual network administration then would be required to complete the move.
Another solution is to have an intelligent exchange of information between the switch and
the devices plugged into the ports. If the switch discovers a device that is trustworthy, it
can extend trust to it dynamically; if not, it will not extend that trust.
Cisco IP phones use the latter solution. In the current Cisco implementation, the intelligent
exchange of information is performed through the Cisco Discovery Protocol (CDP).
Figure 12-6 shows a conditional trust-boundary extension granted to an IP phone that has
passed a CDP exchange.
308
Figure 12-6 Conditionally Trusted Endpoint: IP Phone Trust Boundary Extension and Operation
I see youre an IP phone.

So I will trust your CoS.
IP Phone
Phone VLAN = 110
PC VLAN = 10
TRUST BOUNDARY
4
CoS 5 = DSCP 46
CoS 3 = DSCP 24
CoS 0 = DSCP 0
2
Voice = 5, Signaling = 3
All PC traffic is reset to CoS 0.
PC sets CoS to 5 for all traffic.
1 Switch and phone exchange CDP; trust boundary is extended to IP phone.

2 Phone sets CoS to 5 for VoIP and to 3 for Call-Signaling traffic.
3 Phone rewrites CoS from PC Port to 0.
4 Switch trusts CoS from phone and maps CoS
DSCP for output queuing.
CDP is a lightweight, proprietary protocol engineered to perform neighbor discovery. It

never was intended as a security or authentication protocol. Therefore, to improve the
security of conditional trust extension, the next generation of Cisco IP Telephony products
will incorporate the use of advanced protocols, such as 802.1x and Extensible Authentication
Protocols (EAP), combined with digital certicates, to perform authentication.
Cisco 7902GThe 7902G is an entry-level IP phone that addresses the voicecommunication needs of areas where only a minimal amount of features is required,
such as lobbies, hallways, and break rooms. These phones probably would not be
moved. The 7902G has only a single 10BASE-T Ethernet port on the back of the
phone; therefore, there is no hardware support to connect a PC to it.
Cisco 7905GThe 7905G is a basic IP phone that addresses the voicecommunication needs of a cubicle worker who conducts low to medium telephone
trafc. The Cisco 7905G has only a single 10BASE-T Ethernet port on the back of the
phone; therefore, there is no hardware support to connect a PC to it.
Cisco 7910G and 7910G+SWThe 7910G and 7910G+SW IP phones address

the voice-communication needs associated with a reception area, lab, manufacturing
oor, or employee with a minimal amount of telephone trafc. The only difference
between the Cisco 7910G and the Cisco 7910G+SW is that the former has a single
10BASE-T Ethernet port (therefore, there is no hardware support to connect a PC to
it), and the latter has two 10/100BASE-T Ethernet ports, which allow a PC to be
connected to the IP phone.
309
Cisco 7912GThe 7912G is a basic IP phone that addresses the voice-communication

needs of a cubicle worker who conducts low to medium telephone trafc. The 7912G
supports inline power and an integrated 10/100 Ethernet switch for connecting a PC.
The switch used in the 7912G has the capability to mark CoS and DSCP of Voice and
Call Signaling trafc that originates from the IP phone, but the Cisco 7912G does not
have the capability to re-mark CoS values of PC-generated trafc.
Cisco 7940GThe 7940G IP phone is suited best for an employee in a basic ofce
cubicle environmenta transaction-type worker, for examplewho conducts a
medium amount of business by telephone. The 7940G supports inline power and has
an integrated 10/100 Ethernet switch for connecting a PC.
Cisco 7960GThe 7960G is designed to meet the communication needs of a professional worker in an enclosed ofce environmentan employee who experiences a
high amount of phone trafc in the course of a business day. The 7960G supports
inline power and has an integrated 10/100 Ethernet switch for connecting a PC.
Cisco 7970GThe 7970G not only addresses the needs of the executive or major
decision maker, but also brings network data and applications to users without PCs.
This IP phone includes a backlit, high-resolution color touch-screen display. Currently, Cisco 7970G is the only Cisco IP phone that supports both Cisco prestandard
Power over Ethernet (PoE) and the IEEE 802.3af PoE. The 7970G has an integrated
10/100 Ethernet switch for connecting a PC.
All these IP phones have the capability to mark 802.1Q/p CoS values for both VoIP and Call
Signaling (default values are 5 and 3, respectively). Furthermore, they have the capability
to mark DSCP values for both VoIP and Call Signaling (current defaults are EF and AF31,
respectively; future software releases will change these values to EF and CS3, respectively).
IP phone models 7902G, 7905G, and 7910G lack the hardware to connecting a PC behind
the IP phone. All other IP phone models from the preceding list (except the 7912G) have
the hardware support to connect a PC behind the IP phone and also support 802.1Q/p CoS
re-marking of tagged packets that originate from such PCs.
The 10/100 Ethernet switch built into the 7912G does not have the support to re-mark CoS
values that might have been set by a PC, as illustrated in Figure 12-7. This re-marking
limitation represents a potential security hole for enterprises deploying these IP phones.
However, this hole can be plugged, for the most part, with access-edge policers, as will be
detailed in this chapter. It is important to note that if 7912G IP phones are deployed to users
that move locations, all user switch ports within the enterprise should have access-edge
policers set on them to ensure mobility and security if a 7912G user moves the phones to
another port.
310
Figure 12-7 Conditionally Trusted Endpoint: 7912G IP Phone Trust Boundary Extension and Operation
I see youre an IP phon e.

So I will t rust your CoS.
7912G IP Phone
Phone VLAN = 110
PC VLAN = 10
TRUST BOUNDARY
4
CoS 5 = DSCP 46
CoS 3 = DSCP 24
CoS 5 = DSCP 46
2
Voice = 5, Signaling = 3
No PC traffic is remarked.
PC sets CoS to 5 for all traffic.
1 Switch and phone exchange CDP; trust boundary is extended to IP phone.

2 Phone sets CoS to 5 for VoIP and to 3 for Call-Signaling traffic.
3 Cisco 7912G IP phone does not rewrite CoS from PC port to 0.
4 Switch t rusts CoS from phone and maps CoS
DSCP for output queuing.
Conditionally Trusted IP Phone + PC: Basic Model

In this model, trust (of CoS markings) is extended to CDP-veried IP phones. An additional
layer of protection can be offered by access-edge policers. As stated previously, the tighter
the policers are, the better, provided that adequate bandwidth is permitted for legitimate
applications. The most granular policing can be achieved by the use of per-port or per-VLAN
policers.
NOTE
At the time of this writing, only the Catalyst 3550 family supports per-port and per-VLAN
policing as a feature. Other platforms already have committed to supporting this feature in
the near future. For platforms that do not yet support this feature, equivalent logic can be
achieved by including subnet information within the access lists being referenced by the
class maps. Such examples are provided later in this chapter.
For example, the peak amounts of legitimate trafc originating from the voice VLAN
(VVLAN) are as follows:
128 kbps for Voice trafc (marked CoS 5/DSCP EFthis is 320 kbps, in the case
of G.722 codecs)
311
32 kbps for Call-Signaling trafc (marked CoS 3/DSCP AF31 or CS3)

32 kbps of Best-Effort services trafc (marked CoS 0)
No other trafc should originate from the VVLAN, so the policer could be congured to
re-mark anything else from the VVLAN (because such trafc would be considered
illegitimate and indicative of an attack).
These policers then could be combined with a policer to meter trafc from the data
VLAN (DVLAN), marking down trafc in excess of 5 percent (5 Mbps for FE ports)
to Scavenger/CS1.
Figure 12-8 illustrates the logic of these policers.
Figure 12-8 Conditionally Trusted Endpoint Policing: IP Phone + PC (Basic Model)
Start
VVLAN +
DSCP EF
Yes
No
128 kbps
Yes
Trust and Transmit
No
Drop
VVLAN +
DSCP AF31 or
CS3
Yes
Yes
32 kbps
Yes
5 Mbps
Re-Mark to DSCP CS3

and Transmit
Re-Mark to DSCP CS1
and Transmit
Yes
No
No
DVLAN
ANY
Yes
No
No
VVLAN
ANY
32 kbps
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
Yes
No
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
Conditionally Trusted IP Phone + PC: Advanced Model

Building on the previous model, additional marking and policing can be added for
PC-based videoconferencing and multiple levels of data applications.
312
Desktop videoconferencing applications use the same UDP port range, by default, as does
SoftPhone. If the UDP ports that the desktop videoconferencing application uses can be
dened explicitly within the application, as with SoftPhone, two policers can be used: one
for IP/VC and another for SoftPhone. Otherwise, a single policer covering the UDP port
range of 16,384 to 32,767 is required. This policer would be provisioned for the worst-case
scenario of legitimate trafc. In this case, this would be the videoconferencing applications
requirement of 500 kbps (for a 384-kbps desktop IP/VC application), as compared to
SoftPhones requirement of 128 kbps (or 320 kbps for G.722 codecs).
Additional data VLAN policers can be added to meter Mission-Critical, Transactional, and
Bulk Data ows. Each of these classes can be policed on ingress to the switch port to an
in-prole amount, such as 5 percent each.
NOTE
Because Mission-Critical and Transactional Data applications are interactive foreground

applications that require user input, it is highly unlikely that both of these types of applications will be generating 5 Mbps each simultaneously from a client PC. However, in the rare
case that they are, these ows are policed further by any per-user microow policing
policies that are deployed on distribution-layer Catalyst 6500 Supervisor 720s (PFC3s), as
detailed later in this chapter.
Another factor to keep in mind is that certain Catalyst platforms allow only up to eight
policers per Fast Ethernet port. Therefore, the model presented here is made to conform to
this constraint, to make it more generic and modular. For this reason, a separate policer has
not been dened for Call-Signaling trafc from SoftPhone, but an access list to identify
such trafc could be included within the Mission-Critical Data access lists, which is
detailed in the conguration examples presented later in this chapter.
Figure 12-9 illustrates the logic of these advanced policers.
Figure 12-9 Conditionally Trusted Endpoint Policing: IP Phone + PC (Advanced Model)
VVLAN +
DSCP EF
Start
Yes
Yes
Yes
DVLAN +
MCD ACLs
DVLAN +
TD ACLs
500 kbps
Yes
DVLAN +
BD ACLs
Yes
Yes
Yes
5 Mbps
5 Mbps
No

and Transmit
Re-Mark to DSCP CS1
and Transmit
No
Yes
Re-Mark to DSCP 25
and Transmit
Re-Mark to DSCP CS1
and Transmit
No
Yes

and Transmit
Re-Mark to DSCP CS1
and Transmit
Yes
5 Mbps
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
No
No
DVLAN
ANY
Yes
5 Mbps
Yes
Re-Mark to DSCP CS3

and Transmit
Re-Mark to DSCP CS1
and Transmit
No
No
Bulk
Data ACLs
32 kbps
Yes
No
Transactional
Data ACLs
Yes
No
No
Mission-Critical
Data ACLs
32 kbps
No
No
DVLAN +
UDP 1638432767
Trust and Transmit
Drop
No
VVLAN
ANY
Yes
No
No
VVLAN +
DSCP AF31 or
CS3
128 kbps

and Transmit
Re-Mark to DSCP CS1
and Transmit
Yes
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
313
314
Catalyst 2950 QoS Considerations and Design

The Catalyst 2950 does not support Layer 3 forwarding and, as such, is applicable only as a
(low-end) access-layer switch. Figure 12-10 shows the QoS design options for a Catalyst 2950.
Figure 12-10 Access-Layer Catalyst 2950 QoS Design
Access Edges
Uplinks to
Distribution Layer
Trusted-Endpoint
Model
Untrusted Server
Model
Trust-DSCP
Global 1P3Q1T
Queuing
Conditionally Trusted
PC + IP Phone Basic
Model
It is recommended to use the Enhanced Image (EI) versions of Cisco IOS Software on these
platforms because these offer additional QoS features, such as MQC/ACL classication
options, policing and markdown functions, mapping tables, and AutoQoS.
Catalyst 2950: Trusted Endpoint Model

Conguring a Catalyst 2950 campus to trust an endpoint is fairly straightforward, as shown
in Example 12-1. The trusted endpoint should be assigned either the voice VLAN
(VVLAN) or the data VLAN (DVLAN) with the appropriate switch port commands.
Example 12-1 Catalyst 2950: Trusted Endpoint Example
CAT2950(config)#interface FastEthernet0/1
mls qos trust dscp
Catalyst MLS QoS verication command:
show mls qos interface
Catalyst MLS QoS Verication Command: show mls qos interface

The show mls qos interface verication command reports the congured trust state and the
current operating trust mode of a switch port interface.
315
In Example 12-2, the command veries that interface FastEthernet 0/1 correctly is trusting
the DSCP values of the endpoint to which it is connected.
Example 12-2 show mls qos interface Verication of a Switch Port Connected to a Trusted Endpoint
CAT2950#show mls qos interface FastEthernet0/1
FastEthernet0/1
! Configured trust state is to trust DSCP
trust state: trust dscp
! Current operating mode is to trust DSCP
trust mode: trust dscp
COS override: dis
default COS: 0
pass-through: none
trust device: none
CAT2950#
Catalyst 2950: Untrusted PC with SoftPhone Model

The Catalyst 2950 does not support the range keyword within an ACL when the ACL is
being referenced by an MQC class map. Therefore, a policy to mark UDP ows in the port
range of 16,384 through 32,767 cannot be congured on the Catalyst 2950.
A possible workaround to this limitation is to preset the port(s) to be used by SoftPhone
within the application itself. In such a case, these ports would have to be matched discretely
by ACL entries on the Catalyst 2950. Furthermore, each port being used for Call-Signaling
also would require a discrete ACL entry.
However, even when all these ports are buttoned down and discrete ACLs are congured on
the Catalyst 2950 to match them, another limitation of the switch comes into play. Specifically, the Catalyst 2950 can support policing in only 1-Mbps increments on Fast Ethernet
ports. Such lax policing leaves a fairly large hole to allow unauthorized trafc that is mimicking Voice or Call-Signaling to be admitted onto the network.
Because of these limitations, it is not recommended to use a Catalyst 2950 to support an
untrusted PC running SoftPhone.
Catalyst 2950: Untrusted Server Model

For the most part, the Catalyst 2950 can support the Untrusted Multiapplication Server
model, illustrated in Figure 12-5. Only the nal element of the logical modelnamely, the
policing of all other trafc to 1 Mbps (re-marking trafc in excess of this limit to CS1)is
not supported on the Catalyst 2950.
These main platform-specic caveats should be kept in mind when deploying this model
on the Catalyst 2950:
316
NOTE
Nonstandard DSCP values are not supported; therefore, Mission-Critical Data trafc
cannot be marked to DSCP 25 on Catalyst 2950s (a temporary recommendation
during the interim of the Cisco Call-Signaling marking migration from AF31 to CS3).
Such application trafc alternatively can be marked to the more general class of
Transactional Data (AF21), of which it is a subset.
The mls qos cos override interface command must be used to ensure that untrusted
CoS values explicitly are set to 0 (default).
The range keyword cannot be used in the ACLs being referenced by the class maps;
server ports should be dened explicitly with a separate access control entry (ACE)
per TCP/UDP port.
User-dened masks must be consistent for all ACLs being referenced by class maps.
(If ltering is being done against TCP/UDP ports, all ACEs should be set to lter by
TCP/UDP ports instead of some ACEs ltering by ports and others by subnet or host
addresses.)
System-dened masks (such as permit ip any any) cannot be used in conjunction

with user-dened masks (such as permit tcp any any eq 3200) within the same policy
map. Therefore, if some trafc is being matched against TCP/UDP ports, a nal ACL
cannot be used to match all other trafc through a permit ip any any statement.
The Catalyst 2950 IOS implementation of MQCs class-default currently does not
function compatibly with mainline Cisco IOS. The QoS features and actions dened
within class-default should be applied to all other trafc that is not matched explicitly
by a class map, but testing has shown that this is not the case.
These limitations are based on testing performed using Catalyst 2950 IOS 12.1(19)EA1
versions a through c and 12.1(20)EA1. Additional information on these caveats can be
found at
http://www.cisco.com/univercd/cc/td/doc/product/lan/cat2950/12119ea1/2950scg/swacl.htm
and
http://www.cisco.com/univercd/cc/td/doc/product/lan/cat2950/12119ea1/2950scg/swqos.htm.
Example 12-3 details the Catalyst 2950 conguration to support the Untrusted
Multiapplication Server model (illustrated in Figure 12-5).
Example 12-3 Catalyst 2950: Untrusted Multiapplication Server Example

CAT2950(config)#class-map SAP
CAT2950(config-cmap)# match access-group name SAP
CAT2950(config-cmap)#class-map LOTUS
CAT2950(config-cmap)# match access-group name LOTUS
CAT2950(config-cmap)#class-map IMAP
CAT2950(config-cmap)# match access-group name IMAP
CAT2950(config)#
CAT2950(config)#policy-map UNTRUSTED-SERVER
CAT2950(config-pmap)# class SAP
CAT2950(config-pmap-c)# set ip dscp 18
! DSCP 25 is not supported
CAT2950(config-pmap-c)# police 15000000 8192 exceed-action dscp 8
! Out-of-profile Mission-Critical is marked down to Scavenger (CS1)
CAT2950(config-pmap-c)# class LOTUS
! Transactional Data is marked AF21
! Out-of-profile Transactional Data is marked down to Scavenger (CS1)
CAT2950(config-pmap-c)# class IMAP
! Bulk Data is marked AF11
! Out-of-profile Bulk Data is marked down to Scavenger (CS1)
CAT2950(config-pmap-c)# exit
CAT2950(config-pmap)# exit
CAT2950(config)#
CAT2950(config-if)# mls qos cos override
! Untrusted CoS is remarked to 0
CAT2950(config-if)# service-policy input UNTRUSTED-SERVER
CAT2950(config-if)#exit
CAT2950(config)#
CAT2950(config)#ip access-list extended SAP
CAT2950(config-ext-nacl)# permit tcp any eq 3200 any
CAT2950(config-ext-nacl)#
CAT2950(config-ext-nacl)#ip access-list extended LOTUS
CAT2950(config-ext-nacl)#ip access-list extended IMAP
CAT2950(config-ext-nacl)#end
CAT2950#
317
318
Catalyst MLS QoS verication commands:

show mls qos interface policers
show class-map
show policy-map
show mls masks qos
Catalyst MLS QoS Verication Command: show mls qos interface policers
The show mls qos interface policers verication command reports all congured policers
attached to the specied interface.
In Example 12-4, the policers dened for Mission-Critical, Transactional, and Bulk Data
that are applied to Fast Ethernet 0/1 are conrmed.
Example 12-4 show mls qos interface policers Verication of a Switch Port Connected to an Untrusted
Multiapplication Server
CAT2950#show mls qos interface FastEthernet0/1
FastEthernet0/1
policymap=UNTRUSTED-SERVER
type=Single rate=15000000, burst=8192
!
!
!
CAT2950#
policers
Mission-Critical Data Policer

Transactional Data Policer
Bulk Data Policer
Catalyst MLS QoS Verication Commands: show class-map and show policy-map
The show class-map and show policy-map verication commands report the class map
and policy maps that have been congured globally (regardless of whether theyve been
applied to an interface).
In Example 12-5, the class maps for SAP, LOTUS, and IMAP are displayed, as is the policy
map UNTRUSTED-SERVER that is referencing these.
Example 12-5 show class-map and show policy-map Verication of a Switch Connected to an Untrusted
Multiapplication Server
CAT2950#show class-map
Class Map match-all SAP (id 1)
Match access-group name SAP
Class Map match-all LOTUS (id 2)
Match access-group name LOTUS
Class Map match-all IMAP (id 3)
Match access-group name IMAP
Class Map match-any class-default (id 0)
Match any
319
Example 12-5 show class-map and show policy-map Verication of a Switch Connected to an Untrusted
Multiapplication Server (Continued)

CAT2950#show policy-map
Policy Map UNTRUSTED-SERVER
class SAP
set ip dscp 18
police 15000000 8192 exceed-action dscp 8
class LOTUS
set ip dscp 18
class IMAP
set ip dscp 10
CAT2950#
Catalyst MLS QoS Verication Command: show mls masks qos

The show mls masks qos verication command is helpful in keeping track of the number
of user-dened or system-dened masks that are being applied by ACEs that are referenced
by MQC class maps.
In Example 12-6, the ACEs being referenced by QoS policies are using IP protocol masks,
including (TCP/UDP) source ports.
Example 12-6 show mls masks qos Verication of a Switch Connected to an Untrusted Multiapplication Server
CAT2950#show mls masks qos
Mask1
Type : qos
Fields : ip-proto, src-port
Policymap : UNTRUSTED-SERVER
Interfaces : Fa0/1
CAT2950#

When conguring an access switch to trust or conditionally trust CoS, the default mapping
for CoS 5 should be adjusted to point to DSCP EF (46) instead of DSCP CS5 (40). This
modication is shown in Example 12-7.
Example 12-7 Catalyst 2950: CoS-to-DSCP Marking Modication for Voice
CAT2950(config)#mls qos map cos-dscp 0 8 16 24 32 46 48 56
CAT2950(config)#
! Maps CoS 5 to EF
320
NOTE
Adjusting the default CoS-to-DSCP mapping for Call Signaling (which formerly was
mapped from CoS 3 to DSCP AF31/26) no longer is required. This is because the default
mapping of CoS 3 points to DSCP CS3 (24), which is the Call-Signaling marking that all
Cisco IP Telephony devices markings will migrate to.
show mls qos map

show mls qos map cos-dscp
show mls qos map dscp-cos
Catalyst MLS QoS Verication Command: show mls qos map [cos-dscp | dscp-cos]
The show mls qos map verication command returns the DSCP-to-CoS and CoS-to-DSCP
mappings. These mappings can be either the default mappings or manually congured
overrides.
In Example 12-8, the default mapping for CoS 5 (DSCP CS5) has been modied to point
to DSCP EF instead.
Example 12-8 show mls qos map Verication for a Catalyst 2950 Switch
CAT2950#show mls qos map
Dscp-cos map:
dscp:
0 8 10 16 18 24 26 32 34 40 46 48 56
----------------------------------------------cos:
0 1 1 2 2 3 3 4 4 5 5 6 7
Cos-dscp map:
cos:
0 1 2 3 4 5 6 7
-------------------------------dscp:
0 8 16 24 32 46 48 56
CAT2950#
! CoS 5 is now mapped to DSCP EF
The Catalyst 2950s hardware policers lack the granularity to implement the Conditionally
Trusted IP Phone + PC: Basic model, illustrated in Figure 12-8. However, they can
implement a simplied version of this model, shown in Figure 12-11.
It should be kept in mind that the coarse granularity of the Catalyst 2950s policers (which
are congured in 1 Mbps minimum increments on Fast Ethernet interfaces) potentially
could allow up to 1 Mbps of trafc mimicking legitimate voice trafc per conditionally
trusted switch port.
321
Figure 12-11 Catalyst 2950: Conditionally Trusted Endpoint Policing: IP Phone + PC (Basic Model)
Start
VVLAN
ANY
Yes
1 Mbps
Yes
Trust and Transmit
No
Drop
No
DVLAN
ANY
Yes
5 Mbps
No
Yes
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
Example 12-9 shows the conguration for conguring a switch port to conditionally trust
an IP phone that has a PC connected to it.
Example 12-9 Catalyst 2950: Conditionally Trusted IP Phone + PC: Basic Model Example
CAT2950(config)#mls qos map cos-dscp 0 8 16 24 32 46 48 56 ! Maps CoS 5 to EF
CAT2950(config)#
CAT2950(config)#class-map VVLAN-ANY
CAT2950(config-cmap)# match access-group name VVLAN-ANY
CAT2950(config-cmap)#class-map DVLAN-ANY
CAT2950(config-cmap)# match access-group name DVLAN-ANY
CAT2950(config)#
CAT2950(config)#policy-map IPPHONE+PC
CAT2950(config-pmap)# class VVLAN-ANY
CAT2950(config-pmap-c)#
! Out-of-profile traffic from the VVLAN is dropped
CAT2950(config-pmap-c)# class DVLAN-ANY
set ip dscp 0
! Optional remarking in case trust is compromised
! Out-of-profile data traffic is marked down to Scavenger
CAT2950(config-pmap-c)#exit
CAT2950(config-pmap)#exit
CAT2950(config)#
CAT2950(config)#
CAT2950(config-if)# switchport access vlan 10! DVLAN
CAT2950(config-if)# switchport voice vlan 110! VVLAN
CAT2950(config-if)# mls qos trust device cisco-phone ! Conditional trust
CAT2950(config-if)# mls qos trust cos
! Trust CoS from IP Phone
CAT2950(config-if)# service-policy input IPPHONE+PC ! Policing policy
CAT2950(config)#
CAT2950(config)#ip access-list standard VVLAN-ANY
continues
322
Example 12-9 Catalyst 2950: Conditionally Trusted IP Phone + PC: Basic Model Example (Continued)
CAT2950(config-std-nacl)# permit 10.1.110.0 0.0.0.255
CAT2950(config-std-nacl)#
CAT2950(config-std-nacl)#ip access-list standard DVLAN-ANY
CAT2950(config-std-nacl)# permit 10.1.10.0 0.0.0.255
CAT2950(config-std-nacl)#end
CAT2950#
! VVLAN subnet
! DVLAN subnet

show mls qos map
show class-map
show policy-map
show mls masks qos
Catalyst 2950: Conditionally Trusted IP Phone + PC: Advanced

Model
Because of the previously discussed caveats and limitations of the Catalyst 2950 (including
the maximum number of policers supported per FE interface, the overly coarse policer
granularity, the incapability to mix user-dened masks with system-dened masks, and
other constraints), the Conditionally Trusted IP Phone + PC: Advanced model, shown in
Figure 12-9, cannot be supported on this platform.
Catalyst 2950: Queuing

The Catalyst 2950 can be congured to operate in a 4Q1T mode or in a 1P3Q1T mode (with
queue 4 being congured as a strict-priority queue); the 1P3Q1T mode is recommended for
converged networks.
The strict-priority queue is enabled by conguring the fourth queues weight parameter, as
dened in the wrr-queue bandwidth command, to be 0 (as shown in Example 12-10).
The remaining bandwidth is allocated to the other queues according to their dened
weights. To allocate remaining bandwidths of 5 percent, 25 percent, and 70 percent to
queues 1, 2, and 3, weights of 5, 25, and 70 can be assigned to these queues, respectively.
The logic of these bandwidth allocations recommendations is discussed in more detail
momentarily.
NOTE
323
The absolute values assigned to these queue weights are meaningless, as these weights are
entirely relative. Therefore, these weights can be reduced by dividing each weight by the
lowest common denominator (in this case, 5) to arrive at queue weights of 1, 5, and 14 for
queues 1, 2, and 3, respectively.
Reduction is strictly optional and makes no difference to the servicing of the queues. Many
network administrators tend to prefer dening bandwidth allocation ratios as percentages,
so bandwidth weight ratios are not reduced in this design chapter.
So far, campus QoS designs have been presented for the rst half of the DoS/worm-mitigation
strategy discussed at the beginning of this chapternamely, designs for access-layer
policers to mark down out-of-prole trafc to the Scavenger class PHB of CS1.
The second vital component of this strategy is to map Scavenger class trafc into a lessthan best-effort queuing structure, ensuring that all other trafc will be serviced ahead of it
in the event of congestion.
The Catalyst 2950, like most Catalyst platforms, supports the mapping of CoS values into
queues. The CoS value that corresponds to Scavenger (DSCP CS1) is CoS 1; this CoS value
is shared with Bulk Data (DSCP AF11). Therefore, a small amount of bandwidth (5 percent) is allocated to the less-than-best-effort queue: Q1. Q1 thus services legitimate Bulk
Data trafc but constrains out-of-prole Scavenger trafcwhich could be the result of a
DoS/worm attackto a small amount (less than 5 percent), in the event of congestion.
The next queue, Q2, then is assigned to service Best-Effort trafc. A previously discussed
design principle regarding Best-Effort bandwidth allocation is to allocate approximately
25 percent of a links bandwidth to service Best-Effort trafc. In this manner, the sheer
volume of trafc that defaults to Best-Effort continues to get adequate bandwidth, both in
the event of momentary campus congestion (because of bursts in the amount of legitimate
trafc) and even in the case of a DoS/worm attack.
Preferential applications, such as Transactional Data, Mission-Critical Data, Call-Signaling,
Network and Internetwork Control and Management, and both Interactive- and StreamingVideo, are serviced by Q3. Q3 is allocated 70 percent of the remaining bandwidth (after the
PQ has serviced its Voice trafc).
Figure 12-12 illustrates the recommended 1P3Q1T queuing model for the Catalyst 2950,
along with CoS-to-queue assignments.
324
Figure 12-12 Catalyst 2950 1P3Q1T Queuing Model

Application
DSCP
CoS
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
Interactive-Video
AF41
CoS 4
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
Scavenger
CS1
CoS 1
Best-Effort
1P3Q1T
CoS 5
Q4
Priority Queue
CoS 7
CoS 6
CoS 4
Queue 3
(70%)
CoS 3
CoS 2
CoS 0
Queue 2
(25%)
Queue 1 (5%)
CoS 1
Queue 1 (5%)
The conguration of the priority queue (Q4) and the bandwidth allocations for the
remaining queues (Q1, Q2, and Q3) are shown in Example 12-10.
Example 12-10 Catalyst 2950 Scheduling Conguration: 1P3Q1T Example
CAT2950(config)#wrr-queue bandwidth 5 25 70 0
CAT2950(config)#
! Q1-5%, Q2-25%, Q3-70%, Q4=PQ
show wrr-queue bandwidth
Catalyst MLS QoS Verication Command: show wrr-queue bandwidth

The show wrr-queue bandwidth verication command displays the weights that have
been assigned to the queues. If the command returns a value of 0 for the weight of the fourth
queue, this indicates that the scheduler is operating in 1P3Q1T mode, with Q4 being the
strict-priority queue.
In Example 12-11, the scheduler has been congured for 1P3Q1T queuing: Q1 gets 5 percent
of the remaining bandwidth (after the priority queue has been fully serviced), Q2 gets 25
percent, and Q3 gets 70 percent.
325
Example 12-11 show wrr-queue bandwidth Verication for a Catalyst 2950 Switch
CAT2950#show wrr-queue bandwidth
WRR Queue :
1
2
3
4
5 25 70
0
Bandwidth :
! Q1 gets 5%, Q2 gets 25%, Q3 gets 70%, Q4 is PQ
CAT2950#
Example 12-12 shows the CoS-to-queue mapping conguration for the Catalyst 2950.
Example 12-12 Catalyst 2950 CoS-to-Queue Mapping Example
CAT2950(config)#wrr-queue
CAT2950(config)#
cos-map
cos-map
cos-map
cos-map
1
2
3
4
1
0
2 3 4 6 7
5
!
!
!
!
Scavenger/Bulk is assigned to Q1
Best Effort is assigned to Q2
CoS 2,3,4,6,7 are assigned to Q3
VoIP is assigned to Q4 (PQ)
show wrr-queue cos-map
Catalyst MLS QoS Verication Command: show wrr-queue cos-map

The show wrr-queue cos-map verication command displays the queue to which each
CoS value has been assigned.
In Example 12-13, CoS 0 (Best-Effort) is assigned to Q2 and CoS 1 (Scavenger) has been
assigned to Q1. CoS values 2, 3, 4, 6, and 7 have all been assigned to Q3; CoS 5 (Voice)
has been assigned to the priority queue, Q4.
Example 12-13 show wrr-queue cos-map Verication for a Catalyst 2950 Switch
CAT2950#show wrr-queue cos-map
CoS Value
: 0 1 2 3 4
Priority Queue : 2 1 3 3 3
CAT2950#
5
4
6
3
7
3

The Catalyst 3550 supports IP routing and, thus, can be found in either the access or
distribution layer of the campus.
As for QoS, the Catalyst 3550 supports a richer feature set than the Catalyst 2950, including
an advanced policing feature that is ideal for out-of-prole policing, namely per-port/
per-VLAN policing. The access-layer design options and distribution-layer design recommendations for a Catalyst 3550 are shown in Figures 12-13 and 12-14, respectively.
326

Access Edges
Enable QoS Globally
Trusted-Endpoint Model
1P3Q1T
Queuing
Untrusted
PC + SoftPhone
Model
1P3Q1T
Queuing
1P3Q1T
Queuing
PC + IP Phone
Basic Model
1P3Q1T
Queuing
PC + IP Phone
Advanced Model
1P3Q1T
Queuing
Uplinks to
Distribution Layer
Trust-DSCP
1P3Q2T
Queuing + WRED
Figure 12-14 Distribution-Layer Catalyst 3550 QoS Design
Enable QoS
Globally
Trust-DSCP
1P3Q2T
Queuing + WRED
Interswitch Links
An important point to remember about the Catalyst 3550 is that QoS is disabled by default
and must be enabled globally for congured policies to become effective. While QoS is
disabled, all frames and packets are passed through the switch unaltered (which is
equivalent to a trust CoS and trust DSCP state on all ports). When QoS is enabled globally,
however, all DSCP and CoS values (by default) are set to 0 (which is equivalent to an
untrusted state on all ports). Example 12-14 shows how to verify whether QoS has been
enabled not and also how it can be enabled globally.
Example 12-14 Enabling QoS Globally on the Catalyst 3550
CAT3550#show mls qos
QoS is disabled
CAT3550#
! By default QoS is disabled
327
Example 12-14 Enabling QoS Globally on the Catalyst 3550 (Continued)

CAT3550#configure terminal
Enter configuration commands, one per line. End with CNTL/Z.
mls qos
CAT3550(config)#mls
! Enables QoS globally for the Cat3550
CAT3550(config)#exit
CAT3550#
QoS is enabled
CAT3550#
NOTE
! Verifies that QoS is enabled globally
Depending on the software version, enabling QoS in the Catalyst 3550 might require IEEE
802.3x ow control to be disabled on all interfaces (if it is enabled). Flow control can be
disabled on an interface or interface range by using the interface conguration commands
owcontrol receive off and owcontrol send off. Check the Catalyst 3550 IOS documentation (QoS chapter) to verify whether this is a requirement for the version of software in use.

Conguring a Catalyst 3550 switch port to trust an endpoint is identical to conguring a
Catalyst 2950 (provided that QoS has been enabled globally on the Catalyst 3550). It is
mls qos trust dscp
show mls qos


Unlike the Catalyst 2950, the Catalyst 3550 has all the necessary QoS features to support
and enforce the Untrusted PC with SoftPhone model, as illustrated in Figure 12-4.
Example 12-16 shows the Catalyst 3550 conguration for this access edge model.
Example 12-16 Catalyst 3550: Untrusted PC with SoftPhone Example
CAT3550(config)#mls
mls qos map policed-dscp 0 24 46 to 8
! Excess traffic marked 0 or CS3 or EF will be remarked to CS1
CAT3550(config)#
CAT3550(config)#class-map match-all SOFTPHONE-VOICE
continues
328
Example 12-16 Catalyst 3550: Untrusted PC with SoftPhone Example (Continued)

CAT3550(config-cmap)# match access-group name SOFTPHONE-VOICE
CAT3550(config-cmap)#class-map match-all SOFTPHONE-SIGNALING
CAT3550(config-cmap)# match access-group name SOFTPHONE-SIGNALING
CAT3550(config)#
CAT3550(config)#policy-map SOFTPHONE-PC
CAT3550(config-pmap)#class SOFTPHONE-VOICE
! Softphone VoIP is marked to DSCP EF
CAT3550(config-pmap-c)# police 128000 8000 exceed-action policed-dscp-transmit
! Out-of-profile SoftPhone voice traffic is marked down to Scavenger (CS1)
CAT3550(config-pmap-c)#class SOFTPHONE-SIGNALING
! Signaling is marked to DSCP CS3
! Out-of-profile Signaling traffic is marked down to Scavenger (CS1)
CAT3550(config-pmap-c)#class class-default
! Out-of-profile data traffic is marked down to Scavenger (CS1)
CAT3550(config)#
CAT3550(config-if)# service-policy input SOFTPHONE-PC ! Applies policy to int
CAT3550(config)#
CAT3550(config)#ip access-list extended SOFTPHONE-VOICE
CAT3550(config-ext-nacl)# permit udp any any range 16384 32767 ! VoIP ports
CAT3550(config-ext-nacl)#ip access-list extended SOFTPHONE-SIGNALING
CAT3550(config-ext-nacl)# permit tcp any any range 2000 2002
! SCCP ports
CAT3550#
show mls qos

show mls qos map
show mls qos statistics
show class-map
show policy-map
show policy interface
329
Catalyst MLS QoS Verication Command: show mls qos interface statistics
The show mls qos interface statistics verication command reports dynamic counters for
a given policy, including how many packets were classied and policed by the policy.
In Example 12-17, untrusted packets from the PC are classied and policed according to
the limits shown in Figure 12-4 for the Untrusted PC with SoftPhone access-edge endpoint
policing model.
Example 12-17 show mls qos interface statistics Verication of a Catalyst 3550 Switch Port Connected to an
Untrusted PC with SoftPhone

CAT3550#show mls qos interface FastEthernet0/1 statistics
FastEthernet0/1
Ingress
dscp: incoming
no_change classified policed
dropped (in bytes)
Others: 1275410698 31426318
1243984380 1674978822 0
Egress
dscp: incoming
no_change classified policed
dropped (in bytes)
Others: 7271494
n/a
n/a
0
0
CAT3550#
Catalyst MLS QoS Verication Command: show policy interface

The show policy interface verication command displays the policy maps (and related
classes) that are attached to a given interface.
In Example 12-18, a summary of the untrusted PC with SoftPhone policing policy is shown
as applied to FastEthernet0/1.
Example 12-18 show policy interface Verication of a Catalyst 3550 Switch Port Connected to an Untrusted PC
with SoftPhone
CAT3550#show policy interface FastEthernet0/1
FastEthernet0/1
service-policy input: SOFTPHONE-PC
class-map: SOFTPHONE-VOICE (match-all)
0 packets, 0 bytes
match: access-group name SOFTPHONE-VOICEqm_police_inform_feature:
CLASS_SHOW
class-map: SOFTPHONE-SIGNALING (match-all)
0 packets, 0 bytes
match: access-group name SOFTPHONE-SIGNALINGqm_police_inform_feature:
CLASS_SHOW
class-map: class-default (match-any)
0 packets, 0 bytes
match: any
0 packets, 0 bytes
5 minute rate 0 bpsqm_police_inform_feature: CLASS_SHOW
CAT3550#
330
NOTE
At the time of this writing, the counters reported by the show policy interface command
on the Catalyst 3550 are not being incremented, as is the case with the mainline Cisco IOS
version of this command. This has been reported as a bug. In other words, all counters
currently are frozen at zero; however, when this bug is xed, they should increment dynamically. Catalyst 3550 IOS versions tested and affected with this bug include 12.1(19)EA1 a
through c and 12.1(20)EA1.

The Catalyst 3550 fully supports the Untrusted Multiapplication Server model, depicted in
Figure 12-5. Example 12-19 shows the conguration for this model.
Example 12-19 Catalyst 3550: Untrusted Multiapplication Server Example
CAT3550(config)#mls
mls qos map policed-dscp 0 10 18 25 to 8
! Excess traffic marked 0 or AF11 or AF21 or DSCP 25 will be remarked to CS1
CAT3550(config)#
CAT3550(config)#
CAT3550(config-pmap)#class SAP
! SAP is marked as Mission-Critical
! Out-of-profile SAP is marked down to Scavenger (CS1)
CAT3550(config-pmap-c)#class LOTUS
! Lotus is marked as Transactional
! Out-of-profile LOTUS is marked down to Scavenger (CS1)
CAT3550(config-pmap-c)#class IMAP
! IMAP is marked as Bulk Data
! Out-of-profile IMAP is marked down to Scavenger (CS1)
! Excess data traffic is marked to Scavenger (CS1)
CAT3550(config)#
CAT3550(config)#
331
Example 12-19 Catalyst 3550: Untrusted Multiapplication Server Example (Continued)

CAT3550(config-ext-nacl)# permit tcp any range 3200 3203 any
CAT3550(config-ext-nacl)#permit tcp any eq 1352 any
CAT3550#
show mls qos

show mls qos map
show class-map
show policy-map

The Catalyst 3550s support of per-port and per-VLAN policing gives it a distinct
advantage over other platforms when provisioning a (basic or advanced) Conditionally
Trusted Endpoint model. This is because per-port/per-VLAN policies can be provisioned
without having to enter subnet-specic information for each switch. This makes such
policies more modular and portable.
In Example 12-20, VLAN 10 is the DVLAN and VLAN 110 is the VVLAN.
CAT3550(config)#mls
mls qos map cos-dscp 0 8 16 24 32 46 48 56
! Modifies CoS-to-DSCP mapping to map CoS 5 to DSCP EF
mls qos map policed-dscp 0 24 to 8
CAT3550(config)#mls
! Excess DVLAN & VVLAN traffic will be remarked to Scavenger (CS1)
CAT3550(config)#
CAT3550(config)#
CAT3550(config)#class-map match-all VOICE
continues
332
CAT3550(config-cmap)# match ip dscp 46
! DSCP EF (voice)
CAT3550(config-cmap)#class-map match-any CALL-SIGNALING ! Need 'match-any' here
! DSCP AF31 (old Call-Signaling)
! DSCP CS3 (new Call-Signaling)
CAT3550(config-cmap)#
CAT3550(config-cmap)#class-map match-all VVLAN-VOICE
CAT3550(config-cmap)# match vlan 110
! VLAN 110 is VVLAN
CAT3550(config-cmap)# match class-map VOICE ! Matches VVLAN DSCP EF
CAT3550(config-cmap)#class-map match-all VVLAN-CALL-SIGNALING
! VLAN 110 is VVLAN
CAT3550(config-cmap)# match class-map CALL-SIGNALING !Matches VVLAN AF31/CS3
CAT3550(config-cmap)#class-map match-all ANY
CAT3550(config-cmap)# match access-group name ANY ! Workaround ACL
CAT3550(config-cmap)#class-map match-all VVLAN-ANY
! VLAN 110 is VVLAN
CAT3550(config-cmap)# match class-map ANY
! Matches any other VVLAN traffic
CAT3550(config-cmap)#class-map match-all DVLAN-ANY
! VLAN 10 is DVLAN
! Matches all DVLAN traffic
CAT3550(config-cmap)#policy-map IPPHONE+PC-BASIC
CAT3550(config-pmap)#class VVLAN-VOICE
! DSCP EF (Voice)
CAT3550(config-pmap-c)# police 128000 8000 exceed-action drop
! Only one voice call is permitted per switchport VVLAN
CAT3550(config-pmap-c)#class VVLAN-CALL-SIGNALING
! DSCP CS3 (Call-Signaling)
! Out-of-profile Call-Signaling is marked down to Scavenger (CS1)
CAT3550(config-pmap-c)#class VVLAN-ANY
! Unauthorized VVLAN traffic is marked down to Scavenger (CS1)
CAT3550(config-pmap-c)#class DVLAN-ANY
CAT3550(config)#
CAT3550(config-if)# switchport access vlan 10
! DVLAN
CAT3550(config-if)# switchport voice vlan 110
! VVLAN
CAT3550(config-if)# mls qos trust device cisco-phone
! Conditional Trust
CAT3550(config-if)# service-policy input IPPHONE+PC-BASIC
! Attaches policy
CAT3550(config)#
333
CAT3550(config)#
ip access-list standard ANY
CAT3550(config)#ip
CAT3550(config-std-nacl)# permit any
CAT3550(config-std-nacl)#end
CAT3550#
! Workaround ACL
NOTE
show mls qos

show mls qos map
show class-map
show policy-map
Although Catalyst 3550 IOS syntax supports the match any criteria within a class map
(which the parser allows to be congured in conjunction with a per-VLAN policy), testing
with the Catalyst 3500 IOS versions listed previously has shown that there is a bug with this
function because it does not match any other trafc on a per-VLAN basis. Therefore, an
explicit access list named ANY has been used in Example 12-20 as a workaround to this
issue. Once this issue has been resolved, it is simpler to use the match-any keyword within
the class-maps (rather than the workaround ACL).

The Conditionally Trusted IP Phone + PC: Advanced model builds on the basic model by
including policers for PC-based videoconferencing (and PC SoftPhone), Mission-Critical,
Transactional, and Bulk Data applications. This model is depicted in Figure 12-9. The
Catalyst 3550 can support eight policers per 10/100 Ethernet port and can support this
advanced model.
Example 12-21 shows the Catalyst 3550 conguration for the Conditionally Trusted IP
Phone + PC: Advanced model. In this example, the same server-to-client applications are
used as in the untrusted multiapplication server example. However, notice that the source
and destination ports are reversed for the client-to-server direction of trafc ow. Also,
because of the limit of the number of policers per Fast Ethernet port (eight), there is no
334
explicit policer for SoftPhone Call-Signaling trafc; to work around this limitation,
SoftPhone Call-Signaling trafc is included in the Mission-Critical Data applications
access list.
Example 12-21 Catalyst 3550: Conditionally Trusted IP Phone + PC: Advanced Model Example
CAT3550(config)#mls
mls qos map policed-dscp 0 10 18 24 25 34 to 8
CAT3550(config)#mls
! Excess DVLAN traffic marked 0, AF11, AF21, CS3, DSCP 25,
! and AF41 will be remarked to Scavenger (CS1)
CAT3550(config)#
CAT3550(config)#class-map match-all VOICE
! DSCP EF (voice)
CAT3550(config-cmap)#class-map match-any CALL-SIGNALING ! Need 'match-any' here
CAT3550(config-cmap)#class-map match-all PC-VIDEO
CAT3550(config-cmap)# match access-group name PC-VIDEO
CAT3550(config-cmap)#class-map match-all MISSION-CRITICAL-DATA
CAT3550(config-cmap)# match access-group name MISSION-CRITICAL-DATA
CAT3550(config-cmap)#class-map match-all TRANSACTIONAL-DATA
CAT3550(config-cmap)# match access-group name TRANSACTIONAL-DATA
CAT3550(config-cmap)#class-map match-all BULK-DATA
CAT3550(config-cmap)# match access-group name BULK-DATA
CAT3550(config-cmap)#class-map match-all ANY
CAT3550(config-cmap)# match access-group name ANY ! Workaround ACL
CAT3550(config-cmap)#class-map match-all VVLAN-VOICE
! VLAN 110 is VVLAN
CAT3550(config-cmap)# match class-map VOICE ! Matches VVLAN DSCP EF
! VLAN 110 is VVLAN
CAT3550(config-cmap)# match class-map CALL-SIGNALING ! Matches VVLAN AF31/CS3
! VLAN 110 is VVLAN
CAT3550(config-cmap)#class-map match-all DVLAN-PC-VIDEO
! VLAN 10 is DVLAN
CAT3550(config-cmap)# match class-map PC-VIDEO ! Matches PC-Video class-map
CAT3550(config-cmap)#class-map match-all DVLAN-MISSION-CRITICAL-DATA
! VLAN 10 is DVLAN
CAT3550(config-cmap)# ! Matches MCD class-map
CAT3550(config-cmap)#class-map match-all DVLAN-TRANSACTIONAL-DATA
! VLAN 10 is DVLAN
CAT3550(config-cmap)# match class-map TRANSACTIONAL-DATA ! Matches TD class-map
CAT3550(config-cmap)#class-map match-all DVLAN-BULK-DATA
! VLAN 10 is DVLAN
CAT3550(config-cmap)# ! Matches Bulk Data class-map
335
Example 12-21 Catalyst 3550: Conditionally Trusted IP Phone + PC: Advanced Model Example (Continued)
CAT3550(config-cmap)#class-map match-all DVLAN-ANY
! VLAN 10 is DVLAN
! Matches all other DVLAN traffic
CAT3550(config-cmap)#policy-map IPPHONE+PC-ADVANCED
CAT3550(config-pmap)# class VVLAN-VOICE
! DSCP EF (Voice)
CAT3550(config-pmap-c)#class DVLAN-PC-VIDEO
! DSCP AF41 (Interactive-Video)
! Only one IP/VC stream will be permitted per switchport
CAT3550(config-pmap-c)#class DVLAN-MISSION-CRITICAL-DATA
! Interim Mission-Critical Data
! Out-of-profile Mission-Critical Data is marked down to Scavenger (CS1)
CAT3550(config-pmap-c)#class DVLAN-TRANSACTIONAL-DATA
! DSCP AF21 (Transactional Data)
CAT3550(config-pmap-c)#class DVLAN-BULK-DATA
! DSCP AF11 (Bulk Data)
CAT3550(config-pmap-c)#class DVLAN-ANY
CAT3550(config)#
! DVLAN
! VVLAN
! Conditional Trust
IPPHONE+PC-ADVANCED Attaches policy
CAT3550(config-if)# service-policy input IPPHONE+PC-ADVANCED!
CAT3550(config)#
ip access-list standard ANY ! Workaround ACL
CAT3550(config)#ip
continues
336
CAT3550(config-std-nacl)# permit any
CAT3550(config-std-nacl)#
ip access-list extended PC-VIDEO
CAT3550(config-std-nacl)#ip
! IP/VC or SoftPhone
CAT3550(config-ext-nacl)# permit udp any any range 16384 32767
CAT3550(config-ext-nacl)#ip access-list extended MISSION-CRITICAL-DATA
! SAP
CAT3550(config-ext-nacl)# permit tcp any any eq 3600
! SAP
! SoftPhone SCCP
CAT3550(config-ext-nacl)#ip access-list extended TRANSACTIONAL-DATA
! Lotus
CAT3550(config-ext-nacl)#ip access-list extended BULK-DATA
! IMAP
! IMAP
CAT3550#
show mls qos

show mls qos map
show class-map
show policy-map
Catalyst 3550: Queuing and Dropping

Like the Catalyst 2950, the Catalyst 3550 supports a 1P3Q1T queuing model for all ports.
Gigabit Ethernet ports have the additional option of being congured as 1P3Q2T, with
either tail-drop or WRED thresholds. However, unlike the Catalyst 2950, the Catalyst 3550
queuing parameters are set on a per-interface basis, not globally. Nonetheless, uniform
queuing policies can be deployed expeditiously using the interface range conguration
command.
The strict-priority queue is enabled on a per-interface basis on the Catalyst 3550 with
the priority-queue out interface command. Bandwidth is allocated among the remaining
queues using the wrr-queue bandwidth command. A twist with the Catalyst 3550 is
337
that queue 4s WRR weight is set to 1 (indicating that it does not participate in the WRR
scheduler because it is congured as a strict-priority queue) instead of 0 (as is the case on
the Catalyst 2950). Recommended remaining bandwidth allocations (after the PQ has been
serviced fully) are 5 percent for the Scavenger queue (Q1), 25 percent for the Best-Effort
queue (Q2), and 70 percent for the preferential application queue (Q3).
Following this, CoS 1 (Scavenger/Bulk Data) would be assigned to Q1; CoS 0 (Best-Effort)
would be assigned to Q2; CoS values 2 (Transactional Data and Network-Management),
3 (Call-Signaling and Mission-Critical Data), 4 (Interactive- and Streaming-Video), 6
(Internetwork Control), and 7 (Network Control/Spanning-Tree) would be assigned to Q3;
and CoS 5 (voice) would be assigned to the strict-priority Q4. These assignments and
allocations are illustrated in Figure 12-15 (the thresholds shown in Q1 and Q3 are discussed
shortly).
Application
DSCP
CoS
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
Interactive-Video
AF41
CoS 4
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
1P3Q2T
CoS 5
Q4
Priority Queue
Q3T2
CoS 7
CoS 6
Q3T1
CoS 4
Queue 3
(70%)
CoS 3
CoS 2
Queue 2
(25%)
CoS 0
Scavenger
CS1
CoS 1
Best-Effort
Queue 1 (5%)
CoS 1
Q1T2
Q1T1
Example 12-22 shows the interface-mode conguration commands to congure this

1P3Q1T queuing model, for either Fast Ethernet or Gigabit Ethernet Catalyst 3550
interfaces.
338
Example 12-22 Catalyst 3550 Fast Ethernet and Gigabit Ethernet Interface Queuing Conguration: 1P3Q1T
Example
interface range FastEthernet0/1
CAT3550(config-if)# wrr-queue bandwidth 5 25 70
CAT3550(config-if)# wrr-queue cos-map 1 1
!
!
CAT3550(config-if)# wrr-queue cos-map 3 2 3 4 6
!
CAT3550(config-if)# priority-queue out
!
CAT3550(config-if)#end
CAT3550#
- 48
1
! Q1-5% Q2-25% Q3-70% Q4=PQ
Assigns Scavenger to Q1
Assigns Best Effort to Q2
7 ! Assigns CoS 2,3,4,6,7 to Q3
Assigns VoIP to Q4 (PQ)
Enables Q4 as PQ
show mls qos interface queueing
The Catalyst 3550 offers some advanced nerd-knob queuing options on 10/100 interfaces,
such as tuning minimum reserve thresholds. However, testing has shown that such tuning
makes a highly negligible difference, at best. Therefore, tuning minimum reserve thresholds
is recommended only for advanced network administrators or when using automated tools,
such as AutoQoS.
Some advanced nerd knobs also exist for Gigabit Ethernet interfaces. A couple of these
advanced tuning options include queue-limit tuning and the enabling of WRED thresholds
for 1P3Q2T operation. In the event of DoS or worm attacks, the Gigabit Ethernet uplinks
may become congested, so it is worthwhile to examine these advanced options.
In Example 12-23, the queue limits for both Gigabit Ethernet interfaces are tuned to
correspond to the WRR weights of the queues (the bandwidth allocations). This is achieved
with the wrr-queue queue-limit interface command. However, unlike the WRR weight
bandwidth ratio for Q4 (which is set to 1 to indicate that Q4 is a PQ), the queue limit for
Q4 needs to be set explicitly to a more representative value, such as 30 percent.
NOTE
The default queue limits are such that each queue is assigned 25 percent of the available
buffer space. It should be noted that when the queue limits are modied, the queue
temporarily is shut down during the hardware reconguration, and the switch might drop
newly arrived packets destined to the queue. Thus, it is advisable not to tune the queue
limits on Catalyst 3550 switches already in production networks unless downtime has been
scheduled.
Additionally, WRED is enabled on each (nonpriority) queue. This allows for the preferential
treatment of Bulk Data (DSCP AF11) over Scavenger (CS1) within Q1, as well as the
339
preferential treatment of internetworking or networking protocols (DSCP CS6 and CS7,

respectively) over all other applications assigned to Q3. Even though Q2 has only BestEffort trafc assigned to it, enabling WRED on this queue increases the efciency of TCP
applications within this queue during periods of congestion.
A low WRED threshold, such as 40 percent, can be set for Q1 to aggressively drop Scavenger trafc to preferentially service Bulk Data. The WRED thresholds for Q2 and Q3 can
be set to higher levels, such as 80 percent.
By default, all DSCP values are mapped to the rst WRED threshold of whichever queue
their CoS values are assigned to. Therefore, only DSCP values that are to be mapped to the
second WRED thresholds (of their respective queues) need to be congured manually. In
this case, Bulk Data (DSCP AF11/10), Internetwork Control (DSCP CS6/48), and Network
Control (DSCP CS7/56) all need to be mapped explicitly to the second WRED threshold
using the wrr-queue dscp-map interface conguration command.
NOTE
Network Control trafc in the campus primarily refers to Spanning Tree Protocol (STP)
trafc, such as bridge protocol data units (BPDUs). Although these Layer 2 Ethernet frames
are marked CoS 7, they (obviously) do not have any capabilities to carry Layer 3 DSCP
markings. Thus, it might seem moot to map DSCP CS7 (56) to a higher WRED threshold.
However, it should be kept in mind that Catalyst switches generate internal DSCP values
for all frames (regardless of whether they are carrying IP). These internal DSCP values are
used for QoS decisions, such as WRED, in this case. Therefore, because STP BPDU frames
(marked CoS 7) generate an internal DSCP value of 56, mapping DSCP 56 to the second
threshold of Q3 provides preferential treatment for these important Layer 2 frames.
Example 12-23 shows the conguration for these tuning options, which are available only
on Gigabit Ethernet interfaces on the Catalyst 3550.
Example 12-23 Catalyst 3550 Gigabit Ethernet Interface Queuing and Dropping Conguration: 1P3Q2T
Example
interface range GigabitEthernet 0/1 2
CAT3550(config-if-range)# wrr-queue bandwidth 5 25 70 1
! Q1 gets 5% BW, Q2 gets 25% BW, Q3 gets 70% BW, Q4 is the PQ
CAT3550(config-if-range)# wrr-queue queue-limit 5 25 40 30
! Tunes buffers to 5% for Q1, 25% for Q2, 40% for Q3 and 30% for Q4
40100
100
CAT3550(config-if-range)# wrr-queue random-detect max-threshold 1 5
! Sets Q1 WRED threshold 1 to 40% and threshold 2 to 100%
CAT3550(config-if-range)# wrr-queue random-detect max-threshold 2 80 100
CAT3550(config-if-range)# wrr-queue random-detect max-threshold 3 80 100
CAT3550(config-if-range)# wrr-queue cos-map 11 11
! Assigns Scavenger to Q1
continues
340
Example 12-23 Catalyst 3550 Gigabit Ethernet Interface Queuing and Dropping Conguration: 1P3Q2T
Example (Continued)
cos-map 22 00
CAT3550(config-if-range)# wrr-queue cos-map
! Assigns Best Effort to Q2
CAT3550(config-if-range)# wrr-queue
wrr-queue cos-map
cos-map 33 22 33 44 66 77
! Assigns CoS 2,3,4,6,7 to Q3
cos-map 44 55
CAT3550(config-if-range)# wrr-queue cos-map
! Assigns VoIP to Q4 (PQ)
CAT3550(config-if-range)# wrr-queue dscp-map 2 10 48 56
! Maps Bulk Data (10), Routing (48) and Spanning Tree (Internal DSCP 56)
! to WRED threshold 2 of their respective queues all other DSCP values
! are mapped (by default) to WRED threshold 1 of their respective queues
CAT3550(config-if-range)# priority-queue out
! Enables Q4 as PQ
CAT3550(config-if-range)#end
CAT3550#

show mls qos interface buffers
Catalyst MLS QoS Verication Command: show mls qos interface buffers
The show mls qos interface buffers verication command displays the queue sizes (as perqueue buffer allocation percentages of the total buffer space). Also, the command displays
whether WRED has been enabled on a queue and, if so, displays the rst and second
thresholds (as percentages of the queues depth).
In Example 12-24, the queue limits are set to 5 percent, 25 percent, 40 percent, and 30 percent
of the total queuing buffer space for queues 1 through 4 (respectively). Additionally, WRED
is enabled on queues 1 through 3 (but not Q4 because it is the priority queue). The rst
WRED threshold is set to 5 percent on Q1 and is set to 80 percent on queues 2 and 3.
Example 12-24 show mls qos interface buffers Verication for a Catalyst 3550 Switch
CAT3550#show mls qos interface GigabitEthernet0/1 buffers
GigabitEthernet0/1
Notify Q depth:
qid-size
1 5
! Q1 queue-limit is set to 5% of total buffer space
2 25
3 40
4 30
qid WRED thresh1 thresh2
40
1
ena 5
100
! WRED is enabled on Q1 first threshold is set to 40%
2
ena 80
100
3
ena 80
100
4
dis 100
100
! WRED is disabled on Q4 (as it is the PQ)
CAT3550#
341
Catalyst MLS QoS Verication Command: show mls qos interface queueing
The show mls qos interface queueing verication command displays the CoS-to-queuing
mappings that have been congured, in addition to the bandwidth allocations per queue. On
Gigabit Ethernet interfaces with WRED enabled, the output also includes the DSCP-to-WRED
threshold mappings. This information is displayed in a table form, with the rst digit of the
decimal DSCP value along the y-axis (in rows) and the second digit of the decimal DSCP
value along the x-axis (in columns).
In Example 12-25, the output veries that the egress expedite queue (priority queue, Q4) is
enabled. Also, the WRR bandwidth weights show that, of the remaining bandwidth, Q1
is allocated 5 percent, Q2 is allocated 25 percent, and Q3 is allocated 70 percent.
Additionally, the DSCP-to-WRED table veries that Bulk Data (AF11/10), Internetwork
Control (DSCP CS6/48), and Network Control (DSCP CS7/56) each is mapped to the
second WRED threshold (T2) of its respective queue (as determined by the CoS-to-queue
mappings).
Finally, the CoS-to-queue map shows that CoS 0 (Best-Effort) is assigned to Q2; CoS 1
(Scavenger) has been assigned to Q1; CoS values 2, 3, 4, 6, and 7 have been assigned to
Q3; and CoS 5 (Voice) has been assigned to the priority queue, Q4.
Example 12-25 show mls qos interface queueing Verication for a Catalyst 3550 Switch
CAT3550#show mls qos interface GigabitEthernet0/1 queueing
GigabitEthernet0/1
Egress expedite queue: ena
! Q4 is enabled as a PQ
wrr bandwidth weights:
qid-weights
1 5
! Q1 is allocated 5%
2 25
! Q2 is allocated 25%
3 70
! Q3 is allocated 70&
4 - 1
when expedite queue is disabled
Dscp-threshold map:
d1 : d2 0 1 2 3 4 5 6 7 8 9
--------------------------------------0 :
01 01 01 01 01 01 01 01 01 01
1 :
02 01 01 01 01 01 01 01 01 01
! DSCP 10 is mapped to WRED T2
2 :
01 01 01 01 01 01 01 01 01 01
3 :
01 01 01 01 01 01 01 01 01 01
4 :
01 01 01 01 01 01 01 01 02 01
5 :
01 01 01 01 01 01 02 01 01 01
6 :
01 01 01 01
Cos-queue map:
cos-qid
0 2
! Best-Effort is assigned to Q2
1 1
! Scavenger and Bulk are assigned to Q1
2 3
! Transactional Data and Network Management are assigned to Q3
continues
342
Example 12-25 show mls qos interface queueing Verication for a Catalyst 3550 Switch (Continued)
3 3
4 3
5 4
6 3
7 3
CAT3550#
!
!
!
!
!
Mission-Critical Data and Call-Signaling are assigned to Q3

Interactive- and Streaming-Video are assigned to Q3
Voice is assigned to the priority queue: Q4
Internetwork Control (Routing) is assigned to Q3
Network Control (Spanning Tree) is assigned to Q3
Catalyst 2970/3560/3750 QoS Considerations and

Design
The Catalyst 2970 does not support Layer 3 routing and, as such, is restricted to the role of
an access-layer switch. The 3560 does support Layer 3 routing as well as inline power
a feature that is rarely, if ever, required at the distribution layerand so will only be
considered in an access-layer context. The Catalyst 3750 also supports Layer 3 routing
and may be found in either the access layer or the distribution layer.
Figure 12-16 shows the QoS design options for access-layer Catalyst 2970s, 3560s, or
3750s, and Figure 12-17 shows the QoS design recommendations for a distribution-layer
Catalyst 3750.
Figure 12-16 Access-Layer Catalyst 2950/3560/3750 QoS Design
Access Edges
Enable QoS Globally
1P3Q3T
Queuing + WTD
Untrusted
PC + SoftPhone
Model
1P3Q3T
Queuing + WTD
1P3Q3T
Queuing + WTD
PC + IP Phone
Basic Model
1P3Q3T
Queuing + WTD
PC + IP Phone
Advanced Model
1P3Q3T
Queuing + WTD
Uplinks to
Distribution Layer
Trust-DSCP
1P3Q3T
Queuing + WTD
Catalyst 2970/3560/3750 QoS Considerations and Design
343
Figure 12-17 Distribution-Layer Catalyst 3750 QoS Design
Enable QoS
Globally
1P3Q3T
Queuing + WTD
Trust-DSCP
Interswitch Links
Because (as pointed out in Chapter 10, Catalyst QoS Tools) the QoS features and
conguration syntax are identical for the Catalyst 2970, 3560, and 3750, from a QoS design
recommendation perspective they can be discussed as a single switch.
As with the Catalyst 3550, QoS is disabled globally by default on the Catalyst 2970/3560/
3750. While QoS is disabled, all frames and packets are passed through the switch
unaltered (which is equivalent to a trust CoS and trust DSCP state on all ports). When QoS
is enabled globally, however, all DSCP and CoS values (by default) are set to 0 (which is
equivalent to an untrusted state on all ports).
QoS must be enabled globally for congured policies to become effective. Example 12-26
shows how to verify whether QoS has been enabled and also how it can be enabled globally.
Example 12-26 Enabling QoS Globally on the Catalyst 2970/3560/3750
QoS is disabled
CAT2970#
Enter configuration commands, one per line.
CAT2970(config)#mls
mls qos
CAT2970(config)#end
CAT2970#
End with CNTL/Z.

QoS is enabled
CAT2970#
Catalyst 2970/3560/3750: Trusted Endpoint Model

The Trusted Endpoint model conguration for the Catalyst 2970/3560/3750 is identical to
the conguration of the switches previously discussed (namely, the Catalyst 2950 and
3550); it is shown in Example 12-27.
Example 12-27 Catalyst 2970/3560/3750: Trusted Endpoint Example
CAT2970(config)#interface GigabitEthernet0/1
mls qos trust dscp
344
show mls qos

Catalyst 2970/3560/3750: Untrusted PC with SoftPhone Model

The Untrusted PC with SoftPhone model conguration for the Catalyst 2970/3560/3750
is identical to the Catalyst 3550s for the same access edge model and is shown in
Example 12-28.
Example 12-28 Catalyst 2970/3560/3750: Untrusted PC with SoftPhone Example
mls qos map policed-dscp 0 24 46 to 8
CAT2970(config)#mls
CAT2970(config)#
CAT2970(config)#class-map match-all SOFTPHONE-VOICE
CAT2970(config-cmap)# match access-group name SOFTPHONE-VOICE
CAT2970(config-cmap)#class-map match-all SOFTPHONE-SIGNALING
CAT2970(config-cmap)# match access-group name SOFTPHONE-SIGNALING
CAT2970(config)#
CAT2970(config)#policy-map SOFTPHONE-PC
CAT2970(config-pmap)#class SOFTPHONE-VOICE
CAT2970(config-pmap-c)#class SOFTPHONE-SIGNALING
! Signaling is marked to DSCP CS3
CAT2970(config)#
CAT2970(config-if)# service-policy input SOFTPHONE-PC ! Applies policy to int
CAT2970(config)#
CAT2970(config)#ip access-list extended SOFTPHONE-VOICE
CAT2970(config-ext-nacl)# permit udp any any range 16384 32767 ! VoIP ports
CAT2970(config-ext-nacl)#ip access-list extended SOFTPHONE-SIGNALING
! SCCP ports
CAT2970#
345
show mls qos

show mls qos map
show class-map
show policy-map
Catalyst 2970/3560/3750: Untrusted Server Model

The Untrusted Multiapplication Server model conguration for the Catalyst 2970/3560/
3750 is identical to the conguration of the Catalyst 3550 and is shown in Example 12-29.
Example 12-29 Catalyst 2970/3560/3750: Untrusted Multiapplication Server Example
CAT2970(config)#mls
mls qos map policed-dscp 0 10 18 25 to 8
CAT2970(config)#
CAT2970(config)#
CAT2970(config-pmap)#class SAP
! SAP is marked as Mission-Critical
CAT2970(config-pmap-c)#class LOTUS
! Lotus is marked as Transactional
CAT2970(config-pmap-c)#class IMAP
! IMAP is marked as Bulk Data
! Out-of-profile excess data traffic is marked down to Scavenger (CS1)
continues
346
Example 12-29 Catalyst 2970/3560/3750: Untrusted Multiapplication Server Example (Continued)

CAT2970(config)#
CAT2970(config)#
CAT2970(config-ext-nacl)# permit tcp any range 3200 3203 any
CAT2970#
show mls qos

show mls qos map
show class-map
show policy-map
Catalyst 2970/3560/3750: Conditionally Trusted IP Phone + PC:

Basic Model
The Catalyst 2970/3560/3750 does not support per-port and per-VLAN policing at the time
of this writing. Therefore, access lists are required to match Voice and Call-Signaling trafc
sourced from the VVLAN. These ACLs require the administrator to specify the VVLAN
subnet information. Example 12-30 shows the conguration for a Conditionally Trusted IP
Phone + PC: Basic model for a Catalyst 2970/3560/3750.
Example 12-30 Catalyst 2970/3560/3750: Conditionally Trusted IP Phone + PC: Basic Model Example
CAT2970(config)#mls
CAT2970(config)#mls
mls qos map policed-dscp 0 24 to 8
! Excess VVLAN & DVLAN traffic will be remarked to Scavenger (CS1)
CAT2970(config)#
CAT2970(config)#
347
CAT2970(config)#class-map match-all VVLAN-VOICE
CAT2970(config-cmap)# match access-group name VVLAN-VOICE
CAT2970(config-cmap)# match access-group name VVLAN-CALL-SIGNALING
CAT2970(config-cmap)#policy-map IPPHONE+PC-BASIC
! DSCP EF (Voice)
CAT2970(config)#
CAT2970(config)#
! DVLAN
! VVLAN
! Conditional Trust
CAT2970(config-if)# service-policy input IPPHONE+PC-BASIC
! Attaches policy
CAT2970(config)#
CAT2970(config)#
CAT2970(config)#ip access-list extended VVLAN-VOICE
CAT2970(config-ext-nacl)#permit
permit udp 10.1.110.0 0.0.0.255
any range 16384 32767 dscp ef
! Voice is matched by VVLAN subnet and DSCP EF
CAT2970(config-ext-nacl)#exit
CAT2970(config)#
CAT2970(config)#ip access-list extended VVLAN-CALL-SIGNALING
permit tcp 10.1.110.0 0.0.0.255
any range 2000 2002 dscp af31
! Call-Signaling is matched by VVLAN subnet and DSCP AF31
continues
348
permit tcp 10.1.110.0 0.0.0.255
any range 2000 2002 dscp cs3
! Call-Signaling is matched by VVLAN subnet and DSCP CS3
CAT2970(config)#
CAT2970(config)#ip access-list extended VVLAN-ANY
CAT2970(config-ext-nacl)# permit ip 10.1.110.0 0.0.0.255 any
! Matches all other traffic sourced from the VVLAN subnet
CAT2970#
show mls qos

show mls qos map
show class-map
show policy-map
Catalyst 2970/3560/3750: Conditionally Trusted IP Phone + PC:

Advanced Model
Building on the previous model, PC applications, such as Interactive-Video, Mission-Critical
Data, Transactional Data, and Bulk Data, are identied by access lists. Example 12-31
shows the conguration for the Conditionally Trusted IP Phone + PC Advanced model for
the Catalyst 2970/3560/3750.
CAT2970(config)#mls
CAT2970(config)#mls
! Excess DVLAN traffic marked 0, AF11, AF21, CS3, DSCP 25
CAT2970(config)#
CAT2970(config)#
CAT2970(config)#class-map match-all VVLAN-VOICE
CAT2970(config-cmap)# match access-group name VVLAN-VOICE
CAT2970(config-cmap)# match access-group name VVLAN-CALL-SIGNALING
349
CAT2970(config-cmap)#class-map match-all DVLAN-PC-VIDEO
CAT2970(config-cmap)# match access-group name DVLAN-PC-VIDEO
CAT2970(config-cmap)#class-map match-all DVLAN-MISSION-CRITICAL-DATA
CAT2970(config-cmap)# match access-group name DVLAN-MISSION-CRITICAL-DATA
CAT2970(config-cmap)#class-map match-all DVLAN-TRANSACTIONAL-DATA
CAT2970(config-cmap)# match access-group name DVLAN-TRANSACTIONAL-DATA
CAT2970(config-cmap)#class-map match-all DVLAN-BULK-DATA
CAT2970(config-cmap)# match access-group name DVLAN-BULK-DATA
CAT2970(config)#
CAT2970(config)#policy-map IPPHONE+PC-ADVANCED
! DSCP EF (Voice)
CAT2970(config-pmap-c)#class DVLAN-PC-VIDEO
CAT2970(config-pmap-c)#class DVLAN-MISSION-CRITICAL-DATA
CAT2970(config-pmap-c)#class DVLAN-TRANSACTIONAL-DATA
CAT2970(config-pmap-c)#class DVLAN-BULK-DATA
continues
350
CAT2970(config)#
! DVLAN
! VVLAN
! Conditional Trust
CAT2970(config-if)# service-policy input IPPHONE+PC-ADVANCED!
IPPHONE+PC-ADVANCED Attaches Policy
CAT2970(config)#
CAT2970(config)#
CAT2970(config)#ip access-list extended VVLAN-VOICE
permit udp 10.1.110.0 0.0.0.255
any range 16384 32767 dscp ef
CAT2970(config)#
CAT2970(config)#ip access-list extended VVLAN-CALL-SIGNALING
permit tcp 10.1.110.0 0.0.0.255
any range 2000 2002 dscp af31
permit tcp 10.1.110.0 0.0.0.255
any range 2000 2002 dscp cs3
! Call-Signaling is matched by VVLAN subnet and DSCP AF31 or CS3
CAT2970(config)#
CAT2970(config)#ip access-list extended VVLAN-ANY
CAT2970(config-ext-nacl)# permit ip 10.1.110.0 0.0.0.255 any
CAT2970(config-ext-nacl)#ip access-list extended DVLAN-PC-VIDEO
CAT2970(config-ext-nacl)# permit udp any any range 16384 32767
! IP/VC
CAT2970(config-ext-nacl)#ip access-list extended DVLAN-MISSION-CRITICAL-DATA
! SAP
! SAP
! SCCP
CAT2970(config-ext-nacl)#ip access-list extended DVLAN-TRANSACTIONAL-DATA
! Lotus
CAT2970(config-ext-nacl)#ip access-list extended DVLAN-BULK-DATA
! IMAP
! IMAP
CAT2970#
show mls qos

show mls qos map
351

show class-map
show policy-map
Catalyst 2970/3560/3750: Queuing and Dropping

For the most part, the Catalyst 2970/3560/3750 is relatively compatible in QoS features and
syntax with the Catalyst 3550, except with respect to per-port/per-VLAN policing and
queuing/dropping.
The Catalyst 2970/3560/3750 supports four egress queues, which can be congured on a
per-interface basis to operate in either 4Q3T or 1P3Q3T modes. Additionally, the Catalyst 2970/
3560/3750 supports two queue sets, allowing certain interfaces to be congured in one
manner and others to be congured in a different manner. For example, some interfaces can
be assigned to queue set (qset) 1 operating in 4Q3T mode, while others can be assigned to
queue set 2 operating in 1P3Q3T mode.
However, unlike the Catalyst 2950 and 3550, the Catalyst 2970/3560/3750 has queue 1 (not
queue 4) as the optional priority queue. In a converged campus environment, it is recommended to enable the priority queue using the priority-queue out interface command.
NOTE
The Catalyst 2970/3560/3750 also supports two congurable ingress queues (normal and
expedite). Ingress scheduling, however, is rarely, if ever, required because it becomes
enabled only if the combined input rates from any switch port exceeds the switch fabrics
capacity. Such cases are extremely difcult to achieve, even in controlled lab environments.
In the extreme case that such a scenario should develop in a production environment, the
default settings of the ingress queues are acceptable for maintaining VoIP quality and
network availability.
The three remaining egress queues on the Catalyst 2970/3560/3750 are scheduled by a
shaped round-robin (SRR) algorithm, which can be congured to operate in shaped mode
or in shared mode. In shaped mode, assigned bandwidth is limited to the dened amount;
in shared mode, any unused bandwidth is shared among other classes (as needed).
Shaped or shared bandwidth weights can be assigned to a queue using the srr-queue
bandwidth shape and srr-queue bandwidth share interface commands. Shaped-mode
weights override shared-mode weights. Also, if shaped weights are set to 0, the queue is
operating in shared mode.
352
To make the queuing structure consistent with examples provided for previously discussed
platforms, queues 2 through 4 should be set to operate in shared mode (which is the default
mode of operation on queues 2 through 4). The ratio of the shared weights determines the
relative bandwidth allocations (the absolute values are meaningless). Because the PQ of the
Catalyst 2970/3560/3750 is Q1 (not Q4, as in the Catalyst 3550), the entire queuing model
can be ipped upside down, with Q2 representing the critical data queue, Q3 representing
the Best-Effort queue, and Q1 representing the Scavenger queue. Therefore, shared weights
of 70, 25, and 5 percent can be assigned to queues 2, 3, and 4, respectively.
NOTE
Although the Catalyst 2970/3560/3750 supports the tweaking of queue buffers, testing has
shown that this can interfere with ingress policing policies (on Catalyst IOS versions
12.2[19]EA1ad and 12.2[18]SE for the Catalyst 2970).
Furthermore, adjusting the default thresholds on the Catalyst 3750 sometimes causes
similar interference with ingress policing policies (on Catalyst IOS versions 12.2[19]EA1ad
and 12.2[18]SE for the Catalyst 3750).
Additionally, the Catalyst 2970/3560/3750 supports three weighted tail drop (WTD)
thresholds per queue. Two of these thresholds are congurable (explicit); the third is
noncongurable (implicit) because it is set to the queue-full state (100 percent). These
thresholds can be dened with the mls qos queue-set output qset-id threshold global
command. The only queues that these thresholds need to dene (away from defaults) are
queues 2 and 4. In queue 2, it is recommended to set the rst threshold to 70 percent and the
second to 80 percent, which leaves the third (implicit) threshold set at 100 percent (the tail
of the queue). In queue 4, it is recommended to set the rst threshold to 40 percent, leaving
the default values for both the second and third thresholds at 100 percent.
After the queues and thresholds have been dened, trafc can be assigned to queues and
thresholds by either CoS values or DSCP values, using the mls qos srr-queue output
cos-map queue and mls qos srr-queue output dscp-map queue global commands,
respectively. Although DSCP-to-queue/threshold maps override CoS-to-queue/threshold
maps, these mappings should be as consistent as possible, to ensure predictable behavior
and simplify troubleshooting.
That being said, CoS 0/DSCP 0 (Best-Effort trafc) should be mapped to queue 3 threshold 3
(the tail of the queue) because no other trafc is to be assigned to queue 3.
CoS 1 (Scavenger and Bulk Data) should be mapped to queue 4 threshold 3. Scavenger
trafc can then be contained further by a DSCP-to-queue/threshold mapping assigning
DSCP CS1 to queue 4 threshold 1 (previously set at 40 percent); Bulk Data using DSCP
values AF11, AF12, or AF13 (decimal values 10, 12, and 14, respectively) can use the
remainder of the queue. Bulk Data can use either threshold 2 or threshold 3 as its WTD
limit (both of which are set to 100 percent).
353
CoS 2 and DSCP CS2, AF21, AF22, and AF23 (decimal values 16, 18, 20, and 22, respectively) can be assigned to queue 2 threshold 1 (previously set at 70 percent). This limits
Network-Management and Transactional Data to a subset of queue 2. The temporary marking value for Mission-Critical Data trafc, DSCP 25, also should be assigned to queue 2
threshold 1.
CoS 3, along with DSCP CS3 and AF31 (decimal values 24 and 26, respectively) can be
assigned to queue 2 threshold 2 (previously set to 80 percent). This allows for preferential
treatment of Call-Signaling trafc within queue 2.
CoS 4 and DSCP CS4, AF41, AF42, and AF43 (decimal values 32, 34, 36, and 38, respectively) can be assigned to queue 2 threshold 1. In this manner, video (both interactive and
streaming) will not drown out Call-Signaling or Network and Internetwork Control trafc
within queue 2.
CoS 5 and DSCP EF (decimal value 46) should be assigned to queue 1 threshold 3 because
Voice is the only trafc to be assigned to the strict-priority queue.
CoS 6 and DSCP CS6 (decimal value 48), and CoS 7 and DSCP CS7 (decimal value 56)
should be assigned to queue 2 threshold 3. In this manner, there will always be some room
available in queue 2 to service Network and Internetwork Control trafc.
Figure 12-18 illustrates these recommended Catalyst 2970/3560/3750 assignments from CoS/
DSCP to queues/thresholds.
Figure 12-18 Catalyst 2970/3550 1P3Q3T Queuing Model
Application
DSCP
CoS
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
Interactive-Video
AF41
CoS 4
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
Scavenger
CS1
CoS 1
Best-Effort
1P3Q3T
CoS 1
Queue 4 (5%)
Q4T2
Q4T1
CoS 0
Queue 3
(25%)
Q2T3
CoS 7
CoS 6
CoS 3
Q2T2
Queue 2
(70%)
Q2T1
CoS 4
CoS 2
CoS 5
Q1
Priority Queue
354
Example 12-32 shows the Catalyst 2970/3560/3750 queuing and dropping conguration
recommendations.
Example 12-32 Catalyst 2970/3560/3750: Queuing and Dropping Example
CAT2970(config)#mls
mls qos srr-queue output cos-map queue 1 threshold 3 5
! Maps CoS 5 to Queue 1 Threshold 3 (Voice gets all of Queue 1)
CAT2970(config)#mls
mls qos srr-queue output cos-map queue 2 threshold 1 2 4
! Maps CoS 2 and CoS 4 to Queue 2 Threshold 1
CAT2970(config)#mls
! Maps CoS 3 to Queue 2 Threshold 2
CAT2970(config)#mls
CAT2970(config)#mls
! Maps CoS 0 to Queue 3 Threshold 3 (Best Efforts gets all of Q3)
CAT2970(config)#mls
! Maps CoS1 to Queue 4 Threshold 3 (Scavenger/Bulk gets all of Q4)
CAT2970(config)#
CAT2970(config)#
CAT2970(config)#mls
mls qos srr-queue output dscp-map queue 1 threshold 3 46
! Maps DSCP EF (Voice) to Queue 1 Threshold 3
CAT2970(config)#mls
! Maps DSCP CS2 (Network Management) to Queue 2 Threshold 1
CAT2970(config)#mls
mls qos srr-queue output dscp-map queue 2 threshold 1 18 20 22
! Maps DSCP AF21, AF22, AF23 (Transactional Data) to Queue 2 Threshold 1
CAT2970(config)#mls
! Maps DSCP 25 (Mission-Critical Data) to Queue 2 Threshold 1
CAT2970(config)#mls
! Maps DSCP CS4 (Streaming Video) to Queue 2 Threshold 1
CAT2970(config)#mls
! Maps DSCP AF41, AF42, AF43 (Interactive-Video) to Queue 2 Threshold 1
CAT2970(config)#mls
mls qos srr-queue output dscp-map queue 2 threshold 2 24 26
! Maps DSCP CS3 and DSCP AF31 (Call-Signaling) to Queue 2 Threshold 2
CAT2970(config)#mls
! Maps DSCP CS6 and CS7 (Network/Internetwork) to Queue 2 Threshold 3
CAT2970(config)#mls
! Maps DSCP 0 (Best Effort) to Queue 3 Threshold 3
CAT2970(config)#mls
! Maps DSCP CS1 (Scavenger) to Queue 4 Threshold 1
CAT2970(config)#mls
! Maps DSCP AF11, AF12, AF13 (Bulk Data) to Queue 4 Threshold 3
CAT2970(config)#
CAT2970(config)#
CAT2970(config)#mls
mls qos queue-set output 1 threshold 2 70 80 100 100
! Sets Q2 Threshold 1 to 70% and Q2 Threshold 2 to 80%
CAT2970(config)#mls
mls qos queue-set output 1 threshold 4 5
40100
100100
100100
100
CAT2970(config)#
CAT2970(config)#interface range GigabitEthernet0/1 - 28
CAT2970(config-if-range)# queue-set 1
! Assigns interface to Queue-Set 1 (default)
CAT2970(config-if-range)# srr-queue bandwidth share 1 70 25 5
! Q2 gets 70% of remaining BW; Q3 gets 25% and Q4 gets 5%
355
Example 12-32 Catalyst 2970/3560/3750: Queuing and Dropping Example (Continued)

CAT2970(config-if-range)# srr-queue bandwidth shape 30
! Q1 is limited to 30% of the total available BW
CAT2970(config-if-range)# priority-queue out
CAT2970(config-if-range)#end
CAT2970#
show mls qos interface buffers

show mls qos queue-set
show mls qos maps cos-output-q
show mls qos maps dscp-output-q
Catalyst MLS QoS Verication Command: show mls qos queue-set

The show mls qos queue-set verication command returns the congured buffer
allocations and dened thresholds for each queue set.
In Example 12-33, each queue has a default buffer allocation of 25 percent. Additionally,
all WTD thresholds are set to 100 percent (the tail of the queue), except for queue 2
threshold 1 (set to 70 percent), queue 2 threshold 2 (set to 80 percent), and queue 4
threshold 1 (set to 40 percent).
Example 12-33 show mls qos queue-set Verication for a Catalyst 2970/3560/3750 Switch
CAT2970#show mls qos queue-set 1
Queueset: 1
Queue
:
1
2
3
4
---------------------------------------------buffers
:
25
25
25
25
threshold1:
100
100
40
70
threshold2:
100
100
100
80
reserved :
50
100
50
100
maximum
:
400
100
400
100
CAT2970#
Catalyst MLS QoS Verication Command: show mls qos maps cos-output-q
The show mls qos maps cos-output-q verication command truncates the show mls qos
maps output to report only the CoS-to-queue/threshold mappings for egress queues.
356
In Example 12-34, CoS 0 is mapped to Q3T3, CoS 1 is mapped to Q4T3, CoS 2 is mapped
to Q2T1, CoS 3 is mapped to Q2T2, CoS 4 is mapped to Q2T1, CoS 5 is mapped to Q1T3
(the PQ), and CoS 6 and CoS 7 are mapped to Q2T3.
Example 12-34 show mls qos maps cos-output-q Verication for a Catalyst 2970/3560/3750 Switch
CAT2970#show mls qos maps cos-output-q
Cos-outputq-threshold map:
cos: 0
1
2
3
4
5
6
7
-----------------------------------queue-threshold: 3-3 4-3 2-1 2-2 2-1 1-3 2-3 2-3
CAT2970#
Catalyst MLS QoS Verication Command: show mls qos maps dscp-output-q
The show mls qos maps dscp-output-q verication command truncates the show mls qos
maps output to report only the DSCP-to-queue/threshold mappings for egress queues. The
output is shown in tabular form, with the rst digit of the decimal DSCP value in rows and
the second digit in columns.
In Example 12-35, only standard DSCP PHBs are being mapped away from the default
settings (with the exception of the temporary marking of DSCP 25 for Mission-Critical
Data). The other nonstandard values can be mapped to reect the CoS-to-queue mappings,
but, for example simplicity, this has not been done in this case.
Specically, DSCP 0 is mapped to Q3T3; DSCP CS1 (8) is mapped to Q4T1; DSCP AF11,
AF12, and AF13 (10, 12, 14) are mapped to Q4T3; DSCP CS2 (16) is mapped to Q2T1, as
are DSCP AF21, AF22, and AF23 (18, 20, 22); DSCP CS3 (24) and AF31 (26) are mapped
to Q2T2; DSCP CS4 (32) is mapped to Q2T1, as are DSCP AF41, AF42, and AF43 (34, 36,
38); DSCP EF (46) is mapped to Q1T3 (the PQ); and DSCP CS6 (48) and CS7 (56) are
mapped to Q2T3. The nonstandard DSCP 25 is mapped to Q2T1.
Example 12-35 show mls qos maps dscp-output-q Verication for a Catalyst 2970/3560/3750 Switch
CAT2970#show mls qos maps dscp-output-q
Dscp-outputq-threshold map:
d1 :d2
0
1
2
3
4
5
6
7
8
-----------------------------------------------------------0 :
03-03 02-01 02-01 02-01 02-01 02-01 02-01 02-01 04-01
1 :
04-03 02-01 04-03 02-01 04-03 02-01 02-01 03-01 02-01
2 :
02-01 03-01 02-01 03-01 02-02 02-01 02-02 03-01 03-01
3 :
03-01 03-01 02-01 04-01 02-01 04-01 02-01 04-01 02-01
4 :
01-01 01-01 01-01 01-01 01-01 01-01 01-03 01-01 02-03
5 :
04-01 04-01 04-01 04-01 04-01 04-01 02-03 04-01 04-01
6 :
04-01 04-01 04-01 04-01
CAT2970#
9
02-01
03-01
03-01
04-01
04-01
04-01
Catalyst 4500-SupII+/III/IV/V QoS Considerations and Design
357
Catalyst 4500-SupII+/III/IV/V QoS Considerations and

Design
The Catalyst 4500 with Supervisors II+ through V can be found at either the access layer
or the distribution layer of the campus. Furthermore, because of their high performance,
they also can be found at the core layer of some campus networks.
Figure 12-19 shows the QoS design options for access-layer Catalyst 4500 design; Figure 12-20
shows the distribution- and core-layer recommendations.
Access Edges
Enable QoS Globally
1P3Q1T
Queuing + DBL
Untrusted
PC + SoftPhone
Model
1P3Q1T
Queuing + DBL
1P3Q1T
Queuing + DBL
PC + IP Phone
Basic Model
1P3Q1T
Queuing + DBL
PC + IP Phone
Advanced Model
1P3Q1T
Queuing + DBL
Uplinks to
Distribution Layer
Trust-DSCP
Figure 12-20 Distribution- and Core-Layer Catalyst 4500 QoS Design
Enable QoS
Globally
Trust-DSCP
1P3Q3T
Queuing + WTD
Interswitch Links
1P3Q1T
Queuing + DBL
358
NOTE
To narrow the scope of this discussion to the most current and relevant versions of the
Catalyst 4500 switch family, only the only the Catalyst 4500 with Supervisors II+ through
V is examined is examined in this design chapter. For discussions about older versions of
Catalyst 4000/4500s, refer to the Cisco Press book Cisco Catalyst QoS: Quality of Service
in Campus Networks, by Mike Flannagan, Richard Froom, and Kevin Turek.
Much of the Catalyst MLS QoS syntax is supported on the Catalyst 4500; however, the mls
prex keyword usually is omitted from the conguration commands. For example, as with
the Catalyst 3550 and 2970/3560/3750, QoS is disabled globally on the Catalyst 4500, by
default. However, the command to enable QoS globally on a Catalyst 4500 is simply qos,
not mls qos.
The verication commands are issued in the same manner: with the mls keyword omitted.
Generally, show mls qos [...] verication commands from other Catalyst platforms are
translated to show qos [...] verication commands on the Catalyst 4500 platforms.
Although QoS is disabled globally on the Catalyst 4500, all frames and packets are passed
through the switch unaltered (which is equivalent to a trust CoS and trust DSCP state on all
ports). When QoS is enabled globally, however, all DSCP and CoS values (by default) are
set to 0 (which is equivalent to an untrusted state on all ports).
Example 12-36 shows the verication command to check whether QoS has been globally
enabled on the Catalyst 4500, along with the conguration command to do so.
Example 12-36 Enabling QoS Globally on the Catalyst 4500
CAT4500#show qos
QoS is disabled globally
IP header DSCP rewrite is enabled
CAT4500#
Enter configuration commands, one per line.
CAT4500(config)#qos
qos
CAT4500(config)#end
CAT4500#
CAT4500#show qos
QoS is enabled globally
IP header DSCP rewrite is enabled
CAT4500#
End with CNTL/Z.
359

To enable a given Catalyst 4500 interface to trust the DSCP markings of an endpoint, the
qos trust dscp interface command is used, as shown in Example 12-37.
CAT4500(config-if)# qos trust dscp
CAT4500(config-if)#end
CAT4500#
Catalyst 4500 QoS verication commands:
NOTE
show qos
show qos interface
Because most Catalyst 4500 verication commands are reasonably similar to the MLS QoS
verication commands previously discussed (albeit without the mls keyword), to minimize
redundancy, the MLS QoS verication commands that already have been detailed are not
repeated in this section.

The Untrusted PC with SoftPhone access edge model for Catalyst 4500s, shown in
Example 12-38, is similar to the examples given for previously discussed platforms. A few
distinctions exist, such as the absence of the mls keyword in dening the policed-DSCP
map (along with some slight syntax variation for this command) and the (optional) use of
kbps and mbps (denoting kilobits and megabits, respectively) within the policing
statements.
Example 12-38 Catalyst 4500: Untrusted PC with SoftPhone Model Example
CAT4500-SUP4(config)#qos
qos map dscp policed 0 24 46 to dscp 8
CAT4500-SUP4(config)#
CAT4500-SUP4(config)#class-map match-all SOFTPHONE-SIGNALING
CAT4500-SUP4(config-cmap)# match access-group name SOFTPHONE-SIGNALING
CAT4500-SUP4(config-cmap)#class-map match-all SOFTPHONE-VOICE
CAT4500-SUP4(config-cmap)# match access-group name SOFTPHONE-VOICE
CAT4500-SUP4(config-cmap)#exit
CAT4500-SUP4(config)#policy-map SOFTPHONE-PC
CAT4500-SUP4(config-pmap)# class SOFTPHONE-VOICE
continues
360
Example 12-38 Catalyst 4500: Untrusted PC with SoftPhone Model Example (Continued)
CAT4500-SUP4(config-pmap-c)# set ip dscp ef
CAT4500-SUP4(config-pmap-c)# police 128 kbps 8000 byte exceed-action
policed-dscp-transmit
CAT4500-SUP4(config-pmap-c)#class SOFTPHONE-SIGNALING
CAT4500-SUP4(config-pmap-c)# set ip dscp cs3
! SoftPhone Call-Signaling is marked to DSCP CS3
CAT4500-SUP4(config-pmap-c)#class class-default
CAT4500-SUP4(config-pmap-c)# set ip dscp default
CAT4500-SUP4(config-pmap-c)# police 5 mbps 8000 byte exceed-action
CAT4500-SUP4(config-pmap-c)#exit
CAT4500-SUP4(config)#interface FastEthernet2/1
CAT4500-SUP4(config-if)# service-policy input SOFTPHONE-PC
! Applies policy
CAT4500-SUP4(config-if)#exit
CAT4500-SUP4(config)#ip access-list extended SOFTPHONE-VOICE
CAT4500-SUP4(config-ext-nacl)# permit udp any any range 16384 32767
! VoIP
CAT4500-SUP4(config-ext-nacl)#
CAT4500-SUP4(config-ext-nacl)#ip access-list extended SOFTPHONE-SIGNALING
CAT4500-SUP4(config-ext-nacl)# permit tcp any any range 2000 2002
! SCCP
CAT4500-SUP4(config-ext-nacl)#end
CAT4500-SUP4#
show qos
show qos maps
show qos interface
show class-map
show policy-map

Example 12-39 shows the Catalyst 4500 Untrusted Multiapplication Server model. The
main changes for the Catalyst 4500 for this model are the syntax dening the policed-DSCP
map and the policer denitions (using the abbreviation mbps for megabits per second).
361
Example 12-39 Catalyst 4500: Untrusted Multiapplication Server Model Example

qos map dscp policed 0 10 18 25 to dscp 8
CAT4500-SUP4(config)#class-map SAP
CAT4500-SUP4(config-cmap)# match access-group name SAP
CAT4500-SUP4(config-cmap)#
CAT4500-SUP4(config-cmap)#class-map LOTUS
CAT4500-SUP4(config-cmap)# match access-group name LOTUS
CAT4500-SUP4(config-cmap)#class-map IMAP
CAT4500-SUP4(config-cmap)# match access-group name IMAP
CAT4500-SUP4(config)#policy-map UNTRUSTED-SERVER
CAT4500-SUP4(config-pmap)#class SAP
CAT4500-SUP4(config-pmap-c)# set ip dscp 25
! SAP is marked as Mission-Critical (DSCP 25)
CAT4500-SUP4(config-pmap-c)#class LOTUS
! Lotus is marked as Transactional Data (DSCP AF21)
CAT4500-SUP4(config-pmap-c)#class IMAP
! IMAP is marked as Bulk Data (DSCP AF11)
! Out-of-profile excess data traffic is marked down to Scavenger (CS1)
CAT4500-SUP4(config-pmap-c)# exit
CAT4500-SUP4(config-pmap)#exit
CAT4500-SUP4(config-if)# service-policy input UNTRUSTED-SERVER
CAT4500-SUP4(config)#ip access-list extended SAP
CAT4500-SUP4(config-ext-nacl)# permit tcp any range 3200 3203 any
CAT4500-SUP4(config-ext-nacl)# permit tcp any eq 3600 any
continues
362
Example 12-39 Catalyst 4500: Untrusted Multiapplication Server Model Example (Continued)
CAT4500-SUP4(config-ext-nacl)#ip access-list extended LOTUS
CAT4500-SUP4(config-ext-nacl)#ip access-list extended IMAP
CAT4500-SUP4#
show qos
show qos maps
show qos interface
show class-map
show policy-map

At the time of this writing, the Catalyst 4500 does not support per-port/per-VLAN policing.
Therefore, access lists that include the VVLAN subnet are required to achieve granular
policing of the VVLAN and DVLAN subnets, as shown in Example 12-40.
qos map cos 5 to dscp 46
qos map dscp policed 0 24 to dscp 8
! Excess DVLAN & VVLAN traffic will be marked down to Scavenger (CS1)
CAT4500-SUP4(config)#class-map match-all VVLAN-VOICE
CAT4500-SUP4(config-cmap)# match access-group name VVLAN-VOICE
CAT4500-SUP4(config-cmap)#class-map match-all VVLAN-CALL-SIGNALING
CAT4500-SUP4(config-cmap)# match access-group name VVLAN-CALL-SIGNALING
CAT4500-SUP4(config-cmap)#class-map match-all VVLAN-ANY
CAT4500-SUP4(config-cmap)# match access-group name VVLAN-ANY
CAT4500-SUP4(config-cmap)#policy-map IPPHONE+PC-BASIC
CAT4500-SUP4(config-pmap)#class VVLAN-VOICE
! DSCP EF (Voice)
CAT4500-SUP4(config-pmap-c)# police 128 kbps 8000 byte exceed-action drop
363
CAT4500-SUP4(config-pmap-c)#class VVLAN-CALL-SIGNALING
CAT4500-SUP4(config-pmap-c)#class VVLAN-ANY
CAT4500-SUP4(config-pmap-c)#class
class class-default
CAT4500-SUP4(config-if)# switchport access vlan 10
! DVLAN
CAT4500-SUP4(config-if)# switchport voice vlan 110
! VVLAN
CAT4500-SUP4(config-if)# qos trust device cisco-phone
! Conditional Trust
CAT4500-SUP4(config-if)# service-policy input IPPHONE+PC-BASIC
! MQC Policy
CAT4500-SUP4(config)#ip access-list extended VVLAN-VOICE
CAT4500-SUP4(config-ext-nacl)# permit udp 10.1.110.0 0.0.0.255 any
range 16384 32767
! Voice is matched by VVLAN subnet and UDP port-range
CAT4500-SUP4(config-ext-nacl)#exit
CAT4500-SUP4(config)#ip access-list extended VVLAN-CALL-SIGNALING
CAT4500-SUP4(config-ext-nacl)# permit tcp 10.1.110.0 0.0.0.255 any
range 2000 2002
! Call-Signaling is matched by VVLAN subnet and TCP port-range
CAT4500-SUP4(config-ext-nacl)#exit
CAT4500-SUP4(config)#ip access-list extended VVLAN-ANY
CAT4500-SUP4(config-ext-nacl)# permit ip 10.1.110.0 0.0.0.255 any
CAT4500-SUP4#
show qos
show qos maps
show qos interface
364
show class-map
show policy-map
Catalyst 4500: Conditionally Trusted IP Phone + PC:

Advanced Model
Building on the previous model, PC applications such as Interactive-Video, Mission-Critical
Data, Transactional Data, and Bulk Data are identied by access lists. Example 12-41
shows the conguration for the Conditionally Trusted IP Phone + PC: Advanced model for
the Catalyst 4500.
Example 12-41 Catalyst 4500: Conditionally Trusted IP Phone + PC: Advanced Model Example
qos map cos 5 to dscp 46
qos map dscp policed 0 10 18 24 25 34 to dscp 8
! Excess DVLAN traffic marked 0, AF11, AF21, CS3, DSCP 25
CAT4500-SUP4(config)#class-map match-all VVLAN-VOICE
CAT4500-SUP4(config-cmap)# match access-group name VVLAN-VOICE
CAT4500-SUP4(config-cmap)#class-map match-all VVLAN-CALL-SIGNALING
CAT4500-SUP4(config-cmap)# match access-group name VVLAN-CALL-SIGNALING
CAT4500-SUP4(config-cmap)#class-map match-all VVLAN-ANY
CAT4500-SUP4(config-cmap)# match access-group name VVLAN-ANY
CAT4500-SUP4(config-cmap)#class-map match-all DVLAN-PC-VIDEO
CAT4500-SUP4(config-cmap)# match access-group name DVLAN-PC-VIDEO
CAT4500-SUP4(config-cmap)#class-map match-all DVLAN-MISSION-CRITICAL-DATA
CAT4500-SUP4(config-cmap)# match access-group name DVLAN-MISSION-CRITICAL-DATA
CAT4500-SUP4(config-cmap)#class-map match-all DVLAN-TRANSACTIONAL-DATA
CAT4500-SUP4(config-cmap)# match access-group name DVLAN-TRANSACTIONAL-DATA
CAT4500-SUP4(config-cmap)#class-map match-all DVLAN-BULK-DATA
CAT4500-SUP4(config-cmap)# match access-group name DVLAN-BULK-DATA
CAT4500-SUP4(config)#policy-map IPPHONE+PC-ADVANCED
CAT4500-SUP4(config-pmap)#class VVLAN-VOICE
! DSCP EF (Voice)
CAT4500-SUP4(config-pmap-c)# police 128 kbps 8000 byte exceed-action drop
CAT4500-SUP4(config-pmap-c)#class VVLAN-CALL-SIGNALING
365
CAT4500-SUP4(config-pmap-c)#class VVLAN-ANY
CAT4500-SUP4(config-pmap-c)#class DVLAN-PC-VIDEO
! DSCP AF41 (Int-Video)
CAT4500-SUP4(config-pmap-c)#class DVLAN-MISSION-CRITICAL-DATA
! Interim Mission-Critical
CAT4500-SUP4(config-pmap-c)#class DVLAN-TRANSACTIONAL-DATA
! DSCP AF21
CAT4500-SUP4(config-pmap-c)#class DVLAN-BULK-DATA
! DSCP AF11
CAT4500-SUP4(config-pmap-c)#exit
CAT4500-SUP4(config-if)# switchport access vlan 10
! DVLAN
CAT4500-SUP4(config-if)# switchport voice vlan 110
! VVLAN
CAT4500-SUP4(config-if)# qos trust device cisco-phone
! Conditional Trust
CAT4500-SUP4(config-if)# service-policy input IPPHONE+PC-ADVANCED
! MQC Policy
CAT4500-SUP4(config)#ip access-list extended VVLAN-VOICE
CAT4500-SUP4(config-ext-nacl)# permit udp 10.1.110.0 0.0.0.255 any
range 16384 32767
! Voice is matched by VVLAN subnet and UDP port-range
CAT4500-SUP4(config-ext-nacl)#ip access-list extended VVLAN-CALL-SIGNALING
continues
366
CAT4500-SUP4(config-ext-nacl)# permit tcp 10.1.110.0 0.0.0.255 any
range 2000 2002
! Call-Signaling is matched by VVLAN subnet and TCP port-range
CAT4500-SUP4(config-ext-nacl)#ip access-list extended VVLAN-ANY
CAT4500-SUP4(config-ext-nacl)# permit ip 10.1.110.0 0.0.0.255 any
! VVLAN ANY
CAT4500-SUP4(config-ext-nacl)#ip access-list extended DVLAN-PC-VIDEO
CAT4500-SUP4(config-ext-nacl)# permit udp any any range 16384 32767 ! IP/VC
CAT4500-SUP4(config-ext-nacl)#ip access-list extended DVLAN-MISSION-CRITICAL-DATA
! SAP
CAT4500-SUP4(config-ext-nacl)# permit tcp any any eq 3600
! SAP
! SCCP
CAT4500-SUP4(config-ext-nacl)#ip access-list extended DVLAN-TRANSACTIONAL-DATA
! Lotus
CAT4500-SUP4(config-ext-nacl)#ip access-list extended DVLAN-BULK-DATA
! IMAP
! IMAP
CAT4500-SUP4#
show qos
show qos maps
show qos interface
show class-map
show policy-map
Catalyst 4500: Queuing

The Catalyst 4500 supports four egress queues for scheduling, which can be congured in
either 4Q1T or 1P3Q1T modes. The strict-priority queue on the Catalyst 4500 is transmit
queue 3.
Although tail-drop or WRED thresholds are not supported on the Catalyst 4500, it does
support one of the most advanced congestion-avoidance mechanisms in the Catalyst family.
This congestion-avoidance feature is performed by dynamic buffer limiting (DBL). DBL
tracks the queue length for each trafc ow in the switch and, when the queue length of a
ow exceeds its limit, drops packets or sets the (RFC 3168) explicit congestion notication
(ECN) bits in the IP packet headers. Of course, setting ECN bits is of value only if the
367
end-point applications also support ECN (as discussed in Chapter 6, Congestion-Avoidance

Tools).
DBL can be enabled globally with the qos dbl global command or on a per-class basis
within a policy map with the dbl policy command. A default DBL policy can be applied
to all transmit queues, as shown in the upcoming Example 12-42.
By default, all queues are scheduled in a round-robin manner. The third transmit queue can
be designated as an optional strict-priority queue. This can be enabled as a via the tx-queue 3
interface command, followed by the priority high interface transmit-queue subcommand.
This queue can be dened to be shaped to a peak limit, such as 30 percent, to allow bandwidth
to be available to nonvoice applications. This would be valuable if a trust boundary has been
compromised and a DoS or worm attack is saturating voice queues.
Bandwidth allocations also can be assigned to queues (for certain interfaces) using the
tx-queue interface command, followed by the bandwidth subcommand. Bandwidth
allocations to queues can be assigned only on the following interface types:
Uplink ports on supervisor engines

Ports on the WS-X4306-GB line card
The two 1000BASE-X ports on the WS-X4232-GB-RJ line card
The rst two ports on the WS-X4418-GB line card
The two 1000BASE-X ports on the WS-X4412-2GB-TX line card
The Catalyst 4500 does not support CoS-to-queue mappings; it supports only DSCP-toqueue mappings. These can be dened with the qos map dscp to tx-queue global
command.
Given these features and the objective to make queuing as consistent across platforms as
possible, it is recommended to enable DBL globally on the Catalyst 4500 and to enable
Q3 as the strict-priority queue on all interfaces (so that the switch will operate in 1P3Q1T
mode). This queue can be shaped to 30 percent of the links capacity. Furthermore, Q1 can
be used as the Scavenger/Bulk Data queue, Q2 can be used as the Best-Effort queue, and Q4
can be used as the preferential queue.
On interfaces that support bandwidth allocation, 5 percent can be assigned to Q1, 25 percent
can be assigned to Q2, and 40 percent can be assigned to Q3. Unlike bandwidth weights
that are used on other platforms, these bandwidth allocations are dened in absolute bits
per second or as relative percentages of the links bandwidth. In either case, they should not
total in excess of the links bandwidth limit (1 Gbps, or 100 percent), including the prioritybandwidth allocation for Q3.
By default, the DSCP-to-queue assignments are as follows:
DSCP 0 to 15: queue 1

368

The recommended DSCP-to-queue assignments for the Catalyst 4500 are as follows:
DSCP 0 should be assigned to Q2.
DSCP CS2 (Network-Management) and AF21, AF22, and AF23 (Transactional Data)
should be assigned to Q4.
DSCP CS3 and AF31 (Call-Signaling) should be assigned to Q4.
DSCP EF (Voice) should be assigned to Q3 (the strict-priority queue).
DSCP CS1 (Scavenger) and DSCP AF11, AF12, and AF13 (Bulk Data) should be
assigned to Q1.
DSCP 25 (temporary marking for Mission-Critical Data) should be assigned to Q4.

DSCP CS4 (Streaming-Video) and AF41, AF42, and AF43 (Interactive-Video) should
be assigned to Q4.
DSCP CS6 (Internetwork Control) and CS7 (Network Control/STP) should be
assigned to Q4.
Figure 12-21 illustrates the queuing recommendations for the Catalyst 4500 (Supervisors
II+ through V).
Figure 12-21 Catalyst 4500-SupII+/III/IV/V 1P3Q1T Queuing Model
Application
DSCP
Network Control
(CS7)
CS6
Voice
EF
Interactive-Video
AF41
Streaming-Video
CS4
DSCP 25
Call-Signaling
AF31/CS3
Transactional Data
AF21
Network-Management
CS2
Bulk Data
AF11
1P3Q1T
CS6/CS7
CS4/AF41
Queue 4 (40%)
DSCP 25
CS3/AF31
CS2/AF21
EF
Q3 (30%)
Priority Queue
Queue 2
(25%)
0
Scavenger
CS1
Best-Effort
Queue 1 (5%)
CS1/AF11
Queue 1 (5%)
369
Example 12-42 shows the congurations for enabling queuing on the Catalyst 4500 per
these recommendations. Two separate examples are given: one for a Fast Ethernet interface
that doesnt support bandwidth allocations, and another for a Gigabit Ethernet interface that
does. Some of the DSCP-to-queue mappings shown are not required (because they overlap
with the default settings), but they are shown nonetheless to complete the logic of the
example.
Example 12-42 Catalyst 4500: Queuing and Dropping Examples
qos dbl
! Globally enables DBL
! Optional: Enables DBL to mark RFC 3168 ECN bits in the IP ToS Byte
qos map dscp 0 to tx-queue 2
! Maps DSCP 0 (Best Effort) to Q2
qos map dscp 8 10 12 14 to tx-queue 1
! Maps DSCP CS1 (Scavenger) and AF11/AF12/AF13 (Bulk) to Q1
! Maps DSCP CS2 (Net-Mgmt) and AF21/AF22/AF23 (Transactional) to Q4
qos map dscp 24 25 26 to tx-queue 4
! Maps DSCP CS3 and AF31 (Call-Signaling) and DSCP 25 (MC Data) to Q4
! Maps DSCP CS4 (Str-Video) and AF41/AF42/AF43 (Int-Video) to Q4
! Maps DSCP EF (VoIP) to Q3 (PQ)
qos map dscp 48 56 to tx-queue 4
! Maps DSCP CS6 (Internetwork) and CS7 (Network) Control to Q4
CAT4500-SUP4(config)#policy-map DBL
CAT4500-SUP4(config-pmap)#class class-default
CAT4500-SUP4(config-pmap-c)# dbl
! Enables DBL on all traffic flows
CAT4500-SUP4(config)#interface range FastEthernet2/1 - 48
CAT4500-SUP4(config-if-range)# service-policy output DBL
! Applies DBL policy
CAT4500-SUP4(config-if-range)# tx-queue 3
CAT4500-SUP4(config-if-tx-queue)# priority high
! Enables Q3 as PQ
CAT4500-SUP4(config-if-tx-queue)# shape percent 30
! Shapes PQ to 30%
CAT4500-SUP4(config-if-tx-queue)# exit
CAT4500-SUP4(config-if-range)#exit
CAT4500-SUP4(config)#interface range GigabitEthernet1/1 - 2
CAT4500-SUP4(config-if-range)# service-policy output DBL
CAT4500-SUP4(config-if-range)# tx-queue 1
CAT4500-SUP4(config-if-tx-queue)# bandwidth percent 5
! Q1 gets 5%
CAT4500-SUP4(config-if-tx-queue)# tx-queue 2
! Q2 gets 25%
CAT4500-SUP4(config-if-tx-queue)# priority high
! Enables Q3 as PQ
! PQ gets 30%
CAT4500-SUP4(config-if-tx-queue)# shape percent 30
! Shapes PQ to 30%
continues
370
Example 12-42 Catalyst 4500: Queuing and Dropping Examples (Continued)

CAT4500-SUP4(config-if-tx-queue)#end
CAT4500-SUP4#
! Q4 gets 40%
show qos dbl

show qos maps dscp tx-queue
show qos interface
Catalyst 4500 QoS Verication Command: show qos dbl

The Catalyst 4500 show qos dbl verication command returns whether DBL has been
enabled, along with some of the operating parameters that have been dened for its
operation. These parameters include allowing DBL to set RFC 3168 ECN bits in IP
headers, as shown in Example 12-43.
Example 12-43 show qos dbl Verication for a Catalyst 4500 Switch
CAT4500-SUP4#show qos dbl
QOS is enabled globally
DBL is enabled globally
DBL flow includes vlan
DBL flow includes layer4-ports
DBL uses ecn to indicate congestion
DBL exceed-action probability: 15%
DBL max credits: 15
DBL aggressive credit limit: 10
DBL aggressive buffer limit: 2 packets
CAT4500-SUP4#
Catalyst 4500 QoS Verication Command: show qos maps dscp tx-queue
The Catalyst 4500 show qos maps dscp tx-queue verication command truncates the
show qos maps output to report only the DSCP-to-queue mappings for egress queues. The
output is shown in tabular form, with the rst digit of the decimal DSCP value in rows and
the second digit in columns.
In Example 12-44, only DSCP 0 is mapped to Q2; DSCP CS1 (8) and AF11, AF12, and
AF13 (10, 12, and 14) are mapped to Q1; DSCP CS2 (16) and AF22, AF22, and AF23 (18,
20, and 22) are mapped to Q4; DSCP CS3 (24) and AF31 (26) are mapped to Q4, as is the
nonstandard DSCP 25; DSCP CS4 (32) and AF41, AF42, and AF43 (34, 36, and 38) are
mapped to Q4, as are DSCP CS6 (48) and CS7 (56); and DSCP EF (46) is mapped to Q3.
371
Example 12-44 show qos maps dscp tx-queue Verication for a Catalyst 4500 Switch
CAT4500-SUP4#show qos maps dscp tx-queue
DSCP-TxQueue Mapping Table (dscp = d1d2)
d1 : d2 0 1 2 3 4 5 6 7 8 9
------------------------------------0 :
02 01 01 01 01 01 01 01 01 01
1 :
01 01 01 01 01 01 04 02 04 02
2 :
04 02 04 02 04 04 04 02 02 02
3 :
02 02
4 :
03 03
5 :
04 04
6 :
04 04
CAT4500-SUP4#
04
03
04
04
03 04 03 04 03 04 03
03 03 03 03 03 04 04
04 04 04 04 04 04 04
04
!
!
!
!
!
!
!
!
DSCP
DSCP
DSCP
DSCP
DSCP
DSCP
DSCP
DSCP
0 => Q2; DSCP CS1 => Q1

AF11/AF12/AF13 => Q1
CS2 and AF21 => Q4
AF22/AF23 => Q4
CS3, 25 and AF31 => Q4
CS4 and AF41/AF42/AF43 => Q4
EF => Q3; DSCP CS6 => Q4
CS7 => Q4
Catalyst 4500 QoS Verication Command: show qos interface

The Catalyst 4500 show qos interface verication command displays the global state of
QoS (enabled or not), the trust state of an interface, and any queuing or shaping parameters
that have been dened for the interface.
In Example 12-45, the show qos interface command is being applied on an access-edge
Fast Ethernet interface that has been congured to conditionally trust Cisco IP phones. Furthermore, the output reports that Q3 has been enabled as the priority queue on this interface
and is shaped to 30 Mbps (30 percent). Bandwidth cannot be assigned for nonpriority
queues on this interface, as is indicated by the N/A entries under the bandwidth column.
In Example 12-45, the show qos interface command is being applied to a Gigabit Ethernet
uplink interface that has been congured to trust-DSCP. As before, Q3 has been enabled as
the priority queue and has been shaped to 30 percent, which now translates to 300 Mbps.
Bandwidth is assignable on this interface; therefore, Q1 is allocated 50 Mbps (5 percent),
Q2 is allocated 250 Mbps (25 percent), Q3 is allocated 300 Mbps (30 percent), and Q4 is
allocated 400 Mbps (40 percent).
Example 12-45 show qos interface Verication for a Catalyst 4500 Switch
CAT4500-SUP4#show qos interface FastEthernet2/1
Port QoS is enabled
Administrative Port Trust State: 'cos'
cos
Operational Port Trust State: 'cos'
cos
Trust device: cisco-phone
Default DSCP: 0 Default CoS: 0
Appliance trust: none
Tx-Queue
Bandwidth
ShapeRate
Priority
QueueSize
(bps)
(bps)
(packets)
1
N/A
disabled
N/A
240
2
N/A
disabled
N/A
240
continues
372
Example 12-45 show qos interface Verication for a Catalyst 4500 Switch (Continued)
3
N/A
4
N/A
CAT4500-SUP4#
30000000
disabled
high
N/A
240
240
CAT4500-SUP4#show qos interface GigabitEthernet1/1

Port QoS is enabled
Administrative Port Trust State: 'dscp'
dscp
Operational Port Trust State: 'dscp'
dscp
Trust device: none
Default DSCP: 0 Default CoS: 0
Appliance trust: none
Tx-Queue
Bandwidth
ShapeRate
Priority
QueueSize
(bps)
(bps)
(packets)
1
disabled
N/A
1920
50000000
2
disabled
N/A
1920
250000000
3
1920
300000000
300000000
high
4
disabled
N/A
1920
700000000
CAT4500-SUP4#

The Catalyst 6500 is the undisputed agship of the Cisco family of LAN switches: It is the
most powerful and exible Cisco switching platform. As such, it can be found in all three
layers of a campus network (access, distribution, and core).
NOTE
The Cisco 7600 router is identical in hardware QoS features and conguration to Catalyst
6500s running Cisco IOS. However, the Cisco 7600 is better suited to a MAN or MPLS
VPN environment than a campus context, which is the scope of this chapter. Therefore,
although much of this chapter is applicable to the Cisco 7600, this discussion centers on
Catalyst 6500 QoS.
When congured as an access-layer switch, the preferred software for the Supervisor is
CatOS; when congured as a distribution- or core-layer switch, the preferred software is
Cisco IOS. Figure 12-22 summarizes the QoS design recommendations for a Catalyst 6500
switch at the access layer; Figure 12-23 shows the corresponding recommendations for
Catalyst 6500s deployed in the distribution or core layers.
Figure 12-22 Access-Layer (CatOS) Catalyst 6500 QoS Design

Access Edges
Untrusted
PC + SoftPhone
Model
Enable QoS Globally
Uplinks to
Distribution
Layer
Globally Defined
Linecard-Dependent
Queuing + Dropping
Trust-DSCP
PC + IP Phone
Basic Model
PC + IP Phone
Advanced Model
Figure 12-23 Distribution- or Core-Layer (Cisco IOS) Catalyst 6500 QoS Design
Uplinks from Access Layer Only
Distribution Layer:
Enable QoS
Globally
Trust-DSCP
Optional (PFC3 Only):

Per-User Microflow
Policing
Interface-Group
Line Card-Dependent
Queuing + Dropping
Interswitch Links
Core Layer:
Enable QoS
Globally
Interface-Group
Line Card-Dependent
Queuing + Dropping
Trust-DSCP
Interswitch Links
373
374
NOTE
To narrow the scope of the discussion to the most current and relevant versions of the
Catalyst 6500 switch family, only the Catalyst 6500 with Supervisor 2 (PFC2) and Supervisor 720 (PFC3) is examined in this design chapter. For discussions on older versions of
Catalyst 6000/6500s (such as Supervisor 1, 1a with or without a PFC), refer to the Cisco
Press book Cisco Catalyst QoS: Quality of Service in Campus Networks, by Mike Flannagan,
Richard Froom, and Kevin Turek.
QoS is disabled globally by default on Catalyst 6500s running either CatOS or Cisco IOS.
When QoS is disabled globally, all frames and packets that are passed through the switch
remain unaltered (which is equivalent to a trust CoS and trust DSCP state on all ports).
When QoS is enabled globally, however, all DSCP and CoS values are (by default) set to 0
(which is equivalent to an untrusted state on all ports).
The commands to globally enable and verify QoS on a Catalyst 6500 are shown in
Example 12-46 for CatOS and Example 12-47 for Cisco IOS.
Example 12-46 Enabling QoS Globally on a Catalyst 6500: CatOS
CAT6500-PFC2-CATOS> (enable) set qos enable
QoS is enabled.
CAT6500-PFC2-CATOS> (enable)
CAT6500-PFC2-CATOS> (enable) show qos status
QoS is enabled on this switch.
Example 12-47 Enabling QoS Globally on a Catalyst 6500: Cisco IOS

mls qos
CAT6500-PFC2-IOS(config)#mls
CAT6500-PFC2-IOS(config)#end
CAT6500-PFC2-IOS#
CAT6500-PFC2-IOS#show mls qos
Microflow policing is enabled globally
Vlan or Portchannel(Multi-Earl) policies supported: Yes
----- Module [2] ----QoS global counters:
Total packets: 65
IP shortcut packets: 0
Packets dropped by policing: 0
IP packets with TOS changed by policing: 0
IP packets with COS changed by policing: 0
Non-IP packets with COS changed by policing: 0
CAT6500-PFC2-IOS#
375
Catalyst 6500: CatOS Defaults and Recommendations

CatOS species a number of default QoS settings per port that do not appear in the normal
conguration output. However, it is benecial to be aware of what these defaults are and
what they do, so as not to override them by mistake.
For example, CatOS allows the QoS policy source to be dened by the local conguration
or by the Common Open Policy Source (COPS) protocol, referring to a COPS policydecision point (PDP) Server. COPS is a QoS administration protocol that is both dynamic
and scalable, but, unfortunately, it never gained mainstream acceptance. It is recommended
to leave the switchs default policy source as local (except, of course, in the extremely rare
occurrence that COPS actually is deployed on the network).
Additionally, QoS policies can be applied to VLANs or to ports. There was never any
signicant advantage of using one base over the other, but AutoQoS tools favor port-based
QoS because it is marginally simpler to congure. Port-based QoS is the default per-port
setting, and all examples in this chapter are congured using port-based QoS.
All ports (after QoS has been enabled globally) are set to an untrusted state, by default. The
port CoS setting that all packets are tagged with is 0, by default. Also by default, the trust
extension state is set to untrusted, and the extended-CoS is correspondingly set to 0; this
indicates that any connected IP phones should re-mark PC trafc to 0 on their ASICs.
It is recommended to leave all these port QoS settings at their defaults, with the exception
of trustdepending on the access edge model to be applied, as discussed in the following
section.

For most Catalyst 6500 switch ports, setting the trust state to trust DSCP is a relatively
straightforward command (in either CatOS or Cisco IOS).
In Example 12-48, DSCP-trust is congured on a port in CatOS; in Example 12-51, DSCPtrust is congured on a port/interface in Cisco IOS.
Example 12-48 Catalyst 6500 CatOS: Trusted Endpoint Example
CAT6500-PFC2-CATOS> (enable) set port qos 3/1 trust trust-dscp
Port 3/1 qos set to trust-dscp.
Catalyst 6500 CatOS QoS verication commands:
show qos status

show port qos
376
Catalyst 6500 CatOS QoS Verication Command: show port qos

The Catalyst 6500 CatOS show port qos verication command returns the congured and
runtime QoS states of a port. These might differ because certain commands need to be
committed (programmed into hardware) before they become effective.
In Example 12-49, the switch has QoS globally enabled, and the source of QoS policy
decisions is the local conguration instead of a Common Open Policy Source policydecision point (COPS PDP).
Furthermore, the output shows that the port is congured for port-based QoS (by default)
and has been set to trust DSCP from connected endpoints. No trust extension has been
congured because this is not a conditionally trusted endpoint model.
The output includes the line cards queuing capabilities1P3Q1T (Transmit) and 1P1Q0T
(Receive)along with any ACLs that might be mapped to the port. However, no ACLs have
been mapped to this port in this particular example.
Example 12-49 show port qos Verication for a Catalyst 6500-CatOS Switch
CAT6500-PFC2-CATOS> (enable) show port qos 3/1
QoS is enabled for the switch.
QoS policy source for the switch set to local.
Port Interface Type Interface Type Policy Source Policy Source
config
runtime
config
runtime
----- -------------- -------------- ------------- ------------3/1
port-based
port-based
COPS
local
Port TxPort Type RxPort Type Trust Type
Trust Type
Def CoS Def CoS
config
runtime
config runtime
----- ------------ ------------ ------------ ------------- ------- ------3/1
1p3q1t
1p1q0t
trust-dscp
trust-dscp
0
0
Port Ext-Trust Ext-Cos Trust-Device
----- --------- ------- -----------3/1 untrusted
0
none
(*)Runtime trust type set to untrusted.
Config:
Port ACL name
Type
----- -------------------------------- ---No ACL is mapped to port 3/1.
Runtime:
Port ACL name
Type
----- -------------------------------- ---No ACL is mapped to port 3/1.
On nonGigabit Ethernet line cards that use 2Q2T transmit queuing and 1Q4T receive
queuing, a hardware limitation prevents the proper functioning of port-based trust (which
affects trust-cos, trust-ipprec, and trust-dscp). The show port qos command can be used to
determine whether the line card is a 2Q2T-Tx/1Q4T-Rx line card. These cards also are
listed in Table 12-2.
377
On such line cards, a workaround ACL can be used to achieve trust functionality for trustcos, trust-ipprec, and trust-dscp. Example 12-50 shows the workaround ACL for trustDSCP functionality on such line cards.
Example 12-50 Trust-DSCP Workaround ACL for Catalyst 6500 2Q2T-TX/1Q4T-Rx Non-Gigabit Line Cards
CAT6500-PFC2-CATOS> (enable) set qos acl ip TRUST-DSCP trust-dscp any
TRUST-DSCP editbuffer modified. Use 'commit' command to apply changes.
CAT6500-PFC2-CATOS> (enable) commit qos acl TRUST-DSCP
QoS ACL 'TRUST-DSCP' successfully committed.
CAT6500-PFC2-CATOS> (enable) set qos acl map TRUST-DSCP 4/1
NOTE
To apply the QoS ACL that you have dened (previously), the ACL must be committed to
hardware. The process of committing copies the ACL from a temporary editing buffer to
the PFC hardware. When it is resident in the PFC memory, the policy dened in the QoS
ACL can be applied to all trafc that matches the ACEs. For ease of conguration, most
administrators issue a commit all command. However, you can commit a specic ACL (by
name) to be sent from the editing buffer to PFC memory, as shown in Example 12-50.
In Example 12-51, DSCP-trust is congured on a port/interface in Cisco IOS.

Example 12-51 Catalyst 6500 Cisco IOS: Trusted Endpoint Example
CAT6500-PFC2-IOS(config)#interface FastEthernet3/1
CAT6500-PFC2-IOS(config-if)#mls qos trust dscp
NOTE
show mls qos

The ACL trust workaround to the 2Q2T line card (nonGigabit Ethernet interfaces)
limitation of not supporting trust applies only in the access layer or the campus (where
CatOS is the recommended software for the Catalyst 6500). In the distribution and core
layers, where Cisco IOS is the preferred software, all interfaces are recommended to be
Gigabit Ethernet or higher. Therefore, a Cisco IOS workaround solution for this limitation
of 10/100 2Q2T ports is not warranted.
378

The radical difference in syntax between CatOS and Cisco IOS becomes increasingly
apparent as more complex access edge models are presented.
In the Untrusted PC with SoftPhone model, shown in Example 12-52, per-application
aggregate policers are denedone each for SoftPhone VoIP trafc, SoftPhone CallSignaling trafc, and PC data trafc. Then an ACL (titled SOFTPHONE-PC) with multiple
ACEs is dened, with each ACE referencing its associated aggregate policer. When
complete, the ACL is committed to PFC memory and then mapped to the desired switch
port(s). Switch responses to the commands have been omitted to simplify the example.
Example 12-52 Catalyst 6500 CatOS: Untrusted PC with SoftPhone Model Example
CAT6500-PFC2-CATOS> (enable) set qos policed-dscp-map 0,24,46:8
! Excess traffic marked DSCP 0 or CS3 or EF will be remarked to CS1
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate SOFTPHONE-VOICE
rate 128 burst 8 policed-dscp
! Defines the policer for SoftPhone VoIP traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate SOFTPHONE-SIGNALING
! Defines the policer for SoftPhone Call-Signaling traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate PC-DATA
! Defines the policer for PC Data traffic
CAT6500-PFC2-CATOS> (enable) set qos acl ip SOFTPHONE-PC dscp 46
aggregate SOFTPHONE-VOICE udp any any range 16384 32767
! Binds ACL to policer and marks in-profile SoftPhone VoIP to DSCP EF
aggregate SOFTPHONE-SIGNALING tcp any any range 2000 2002
! Binds ACL to policer marks in-profile Call-Signaling to DSCP CS3
aggregate PC-DATA any
! Binds ACL to policer and marks in-profile PC Data traffic to DSCP 0
CAT6500-PFC2-CATOS> (enable) commit qos acl SOFTPHONE-PC
! Commits ACL to PFC memory
CAT6500-PFC2-CATOS> (enable) set port qos 3/1 trust untrusted
! Sets the port trust state to untrusted
CAT6500-PFC2-CATOS> (enable) set qos acl map SOFTPHONE-PC 3/1
! Attaches ACL to switch port
show qos status

show qos maps
show port qos
379
show qos acl

show qos policer
show qos statistics
Catalyst 6500 CatOS QoS Verication Command: show qos maps

The Catalyst 6500 CatOS show qos maps verication command is fairly similar to the
show mls qos maps MLS QoS verication command. It returns the congured CoS-DSCP,
IPPrec-DSCP, DSCP-CoS, and Normal-Rate and Excess-Rate Policed-DSCP maps. The
command can return congured maps (which might or might not be committed to the PFC)
or runtime maps.
In the (truncated) runtime example shown in Example 12-53, all maps are at their default
states, with the exception of the Normal-Rate Policed-DSCP map, which has DSCP 0, CS3
(24), and EF (46) mapped for out-of-prole markdown to DSCP CS1 (8).
Example 12-53 show qos maps Verication for a Catalyst 6500-CatOS Switch
CAT6500-PFC2-CATOS> (enable) show qos maps runtime
CoS - DSCP map:
CoS
DSCP
-----0
0
1
8
2
16
3
24
4
32
5
40
6
48
7
56
IP-Precedence - DSCP map:
IP-Prec
DSCP
---------0
0
1
8
2
16
3
24
4
32
5
40
6
48
7
56
DSCP - CoS map:
DSCP
-------------------------------0-7
8-15
16-23
CoS
--0
1
2
continues
380
Example 12-53 show qos maps Verication for a Catalyst 6500-CatOS Switch (Continued)
24-31
32-39
40-47
48-55
56-63
3
4
5
6
7
DSCP - Policed DSCP map normal-rate:

DSCP
Policed DSCP
-------------------------------- -----------1 1
2 2
3 3
4 4
5 5
6 6
7 7
0,8,24,46 8
9 9
10 10
<output truncated>
63 63
DSCP - Policed DSCP map excess-rate:
DSCP
Policed DSCP
-------------------------------- -----------0 0
1 1
2 2
3 3
4 4
5 5
<output truncated>
63 63
Catalyst 6500 CatOS QoS Verication Command: show qos acl

The Catalyst 6500 CatOS show qos acl verication command returns information about
ACLs and ACEs that have been congured for QoS purposes. ACL information can be
displayed for conguration ACLs or runtime ACLs.
Example 12-54 displays three variations of the show qos acl command.
The rst one displays the QoS ACLs that are still in the edit buffer and indicates whether
the ACLs have been committed to PFC memory. In this example, the ACL SOFTPHONE-PC
has been committed to the PFC.
The second example displays runtime ACE-level information for a given ACL (or all ACLs,
if the keyword all is used instead of the ACL name). Each ACEs DSCP markings,
associated aggregate policer, and ltering criteria are displayed.
381
The third example displays the VLANs and ports that the ACL has been applied to. In this
specic example, port 3/1 has the SOFTPHONE-PC ACL applied to it.
Example 12-54 show qos acl Verication for a Catalyst 6500-CatOS Switch
CAT6500-PFC2-CATOS> (enable) show
ACL
-------------------------------SOFTPHONE-PC
qos acl editbuffer

Type Status
---- ---------IP
Committed
CAT6500-PFC2-CATOS> (enable) show qos acl info runtime SOFTPHONE-PC

set qos acl IP SOFTPHONE-PC
---------------------------------------------1. dscp 46 aggregate SOFTPHONE-VOICE udp any any range 16384 32767
2. dscp 24 aggregate SOFTPHONE-SIGNALING tcp any any range 2000 2002
3. dscp 0 aggregate PC-DATA any
CAT6500-PFC2-CATOS> (enable) show qos

QoS ACL mappings on input side:
ACL name
Type
-------------------------------- ---SOFTPHONE-PC
IP
ACL name
Type
-------------------------------- ---SOFTPHONE-PC
IP
acl map runtime SOFTPHONE-PC

Vlans
--------------------------------Ports
--------------------------------3/1
Catalyst 6500 CatOS QoS Verication Command: show qos policer

The Catalyst 6500 CatOS show qos policer verication command displays the normal and
excess rates and burst for policers.
In Example 12-55, three aggregate policers have been dened: SOFTPHONE-VOICE,
SOFTPHONE-SIGNALING, and PC-DATA, with normal rates of 128 kbps, 32 kbps, and
5 Mbps, respectively. Each of these policers marks down excess trafc according to the
policed-DSCP normal rate and are attached to the ACL SOFTPHONE-PC.
Example 12-55 show qos policer Verication for a Catalyst 6500-CatOS Switch
CAT6500-PFC2-CATOS> (enable) show qos policer runtime all
Warning: Runtime information may differ from user configured setting due to
hardware granularity.
QoS microflow policers:
QoS aggregate policers:
Aggregate name
Avg. rate (kbps) Burst size (kb) Normal action
--------------------- ---------------- --------------- -------------
continues
382
Example 12-55 show qos policer Verication for a Catalyst 6500-CatOS Switch (Continued)
SOFTPHONE-VOICE
128
8
policed-dscp
Excess rate (kbps) Excess burst size (kb) Excess action
------------------ ---------------------- ------------31457280
31744 policed-dscp
ACL attached
-----------------------------------SOFTPHONE-PC
Aggregate name
--------------------- ---------------- --------------- ------------SOFTPHONE-SIGNALING
32
8 policed-dscp
------------------ ---------------------- ------------31457280
31744 policed-dscp
ACL attached
-----------------------------------SOFTPHONE-PC
Aggregate name
--------------------- ---------------- --------------- ------------PC-DATA
4864
8
policed-dscp
------------------ ---------------------- ------------31457280
31744 policed-dscp
ACL attached
-----------------------------------SOFTPHONE-PC
Catalyst 6500 CatOS QoS Verication Command: show qos statistics

The Catalyst 6500 CatOS show qos statistics verication command displays various
dynamic statistics regarding the QoS policies.
In the three parts of Example 12-56, the rst variation of the command show qos statistics
mod | port returns queuing statistics for the port. Specically, it reports any drops because
of queue buffer overll and breaks down these drops by queues or thresholds, depending
on the queuing structure of the module. In this rst part of the example, no drops have
occurred because of queuing buffer overlls.
In the second part of Example 12-56, aggregate policing statistics are displayed through the
show qos statistics aggregate-policer variation of the command. The number of packets
that conform to or exceed a given policer is reported. In this part of the example, the
command reports that no packets have exceeded the SOFTPHONE-VOIP or
SOFTPHONE-SIGNALING policers, but a few have exceeded the PC-DATA policer.
Finally, in the third part of Example 12-56, the number of packets that have been dropped
because of policing or have been re-marked at Layer 3 or Layer 2 is reported with the show
qos statistics l3stats command. In this nal part of the example, the command reports that
no packets have been dropped because of policing, but thousands of packets have had their
L3 and L2 markings modied by the congured policy.
383
Example 12-56 show qos statistics Verication for a Catalyst 6500-CatOS Switch
CAT6500-PFC2-CATOS> (enable) show qos statistics 3/1
Tx port type of port 3/1 : 1p3q1t
WRED and tail drops are accumulated in one counter per queue.
Q # Packets dropped
--- ----------------------------------------------1
0 pkts
2
0 pkts
3
0 pkts
4
0 pkts
Rx port type of port 3/1 : 1p1q0t
For untrusted ports all the packets are sent to the same queue,
Rx thresholds are disabled, tail drops are reported instead.
Q # Threshold #:Packets dropped
--- ----------------------------------------------1
0:0 pkts
2
0:0 pkts
CAT6500-PFC2-CATOS> (enable) show qos statistics aggregate-policer

QoS aggregate-policer statistics:
Aggregate policer
Allowed packet Packets exceed
count
excess rate
------------------------------- -------------- -------------SOFTPHONE-VOICE
27536
0
SOFTPHONE-SIGNALING
224
0
PC-DATA
470069
645
CAT6500-PFC2-CATOS>
Packets dropped due
IP packets with ToS
IP packets with CoS
Non-IP packets with
CAT6500-PFC2-CATOS>
(enable) show qos statistics l3stats

to policing:
0
changed:
169286
changed:
83507
CoS changed:
0
(enable)

Additional exibility is offered to the Untrusted Server model with the Catalyst 6500 PFC2/
PFC3s support of dual-rate policing (as described in RFC 2698, A Two Rate Three Color
Marker, and as illustrated in Figure 4-5).
Using a dual-rate policer, three colors are used to indicate the following:
Conforming trafc (within the normal rate)

Excess trafc (exceeding the normal rate but less than the excess rate)
Violating trafc (exceeding both the normal and excess rates)
384
The dual-rate policer is intended to complement the RFC 2597 assured-forwarding groups
DiffServ marking scheme. To illustrate this, consider Transactional Data trafc, which is
marked to AF Class 2. Conforming Transactional Data should be marked to AF21, excess
Transactional Data trafc should be marked down to AF22, and violating Transactional
Data trafc should be marked down further to AF23.
Such a markdown scheme is intended to be complemented further by DSCP-based WRED
congestion avoidance. In this manner, if congestion occurs, AF23 is dropped more aggressively than AF22, which, in turn, is dropped more aggressively than AF21.
However, because Catalyst 6500 queuing and congestion avoidance are determined primarily by CoS markings, the standards-based DSCP model cannot be followed completely at
this time on this platform. (Because AF21, AF22, and AF23 all share CoS 3, this does not
allow for granular subclass QoS.) Therefore, Scavenger class markings for violating trafc
could be used to achieve a similar overall effect, while maintaining consistency with QoS
designs previously presented for other Catalyst platforms.
Under such a modied Untrusted Multiapplication Server model, excess Transactional
Data trafc can be marked down to AF22 and violating Transactional Data trafc can be
marked down to DSCP CS1 (Scavenger). Similarly, excess Bulk Data trafc can be marked
down to AF12 and violating Bulk Data trafc can be marked down to DSCP CS1
(Scavenger). This modied model is shown in Figure 12-24.
Figure 12-24 Catalyst 6500 PFC2/PFC3 Untrusted Endpoint Dual-Rate Policing: Multiapplication Server
Example
Start
Mission-Critical
Data ACLs
SAP ACLs
15 Mbps
Yes
No
No
Lotus ACLs
Transactional
Data ACLs
Yes
Yes
25 Mbps
No
No
No
Re-Mark to DSCP CS1

and Transmit
Yes
Yes

and Transmit
and Transmit
35 Mbps
Re-Mark to DSCP CS1
and Transmit
No
IMAP
ACLs
Bulk
Data ACLs
Re-Mark to DSCP 25
and Transmit
Yes
40 Mbps
No
Yes

and Transmit
Yes

and Transmit
No
50 Mbps
Re-Mark to DSCP CS1

and Transmit
No
1 Mbps
No
Yes
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
385
Example 12-57 shows the conguration for a Catalyst 6500 CatOS untrusted endpoint dualrate policing of a multiapplication server.
Example 12-57 Catalyst 6500 CatOS: Untrusted Multiapplication Server Example
(Dual-Rate Policing)
CAT6500-PFC2-CATOS> (enable) set qos policed-dscp-map normal-rate 0,25:8
! Excess SAP and Data traffic is marked down to DSCP CS1 (Scavenger)
CAT6500-PFC2-CATOS> (enable) set qos policed-dscp-map normal-rate 18:20
! Excess Transactional Data traffic is marked down from DSCP AF21 to AF22
CAT6500-PFC2-CATOS> (enable) set qos policed-dscp-map excess-rate 18:8
! Violating Transactional Data traffic is marked down to CS1
! Excess Bulk Data traffic is marked down from DSCP AF11 to AF12
! Violating Bulk Data traffic is marked down to CS1
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate SAP
! Defines the policer for Mission-Critical Data (SAP) traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate LOTUS
rate 25000 policed-dscp erate 35000 policed-dscp burst 8
! Defines the dual-rate policer for Transactional Data (Lotus) traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate IMAP
! Defines the dual-rate policer for Bulk Data (IMAP) traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate DATA
! Defines the policer for other data traffic
CAT6500-PFC2-CATOS> (enable) set qos acl ip UNTRUSTED-SERVER dscp 25
aggregate SAP tcp any range 3200 3203 any
! Binds ACL to policer and marks in-profile SAP to DSCP 25
aggregate SAP tcp any eq 3600 any
aggregate LOTUS tcp any eq 1352 any
! Binds ACL to dual-rate policer and marks in-profile Lotus to DSCP AF21
aggregate IMAP tcp any eq 143 any
! Binds ACL to dual-rate policer and marks in-profile IMAP to DSCP AF11
aggregate IMAP tcp any eq 220 any
! Binds ACL to dual-rate policer and marks in-profile IMAP to DSCP AF11
aggregate DATA any
continues
386
Example 12-57 Catalyst 6500 CatOS: Untrusted Multiapplication Server Example
(Dual-Rate Policing) (Continued)

! Sets the port trust
commit qos acl UNTRUSTED-SERVER

set port qos 3/1 trust untrusted
state to untrusted
set qos acl map UNTRUSTED-SERVER 3/1
show qos status

show qos maps
show port qos
show qos acl
show qos policer
show qos statistics

In the Conditionally Trusted IP Phone + PC model for the Catalyst 6500 (CatOS), four
aggregate policers are dened, one each for Voice from the VVLAN, Call Signaling from
the VVLAN, all other trafc from the VVLAN, and all PC data trafc. Conditional trust is
extended to the IP phones using the trust-device command, as shown in Example 12-58.
Example 12-58 Catalyst 6500 CatOS: Conditionally Trusted IP Phone + PC: Basic Model Example
CAT6500-PFC2-CATOS> (enable) set qos cos-dscp-map 0 8 16 24 32 46 48 56
! Modifies default CoS-DSCP mapping so that CoS 5 is mapped to DSCP EF
CAT6500-PFC2-CATOS> (enable) set qos policed-dscp-map 0,24:8
! Excess traffic marked DSCP 0 or CS3 is remarked to CS1
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate VVLAN-VOICE
! Defines the policer for IP Phone VoIP traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate VVLAN-SIGNALING
! Defines the policer for IP Phone Call-Signaling traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate VVLAN-ANY
! Defines the policer for any other traffic sourced from the VVLAN
! Defines the policer for PC Data traffic
CAT6500-PFC2-CATOS> (enable) set qos acl ip IPPHONE-PC-BASIC dscp 46
aggregate VVLAN-VOICE udp 10.1.110.0 0.0.0.255 any range 16384 32767
! Binds ACL to policer and marks in-profile VVLAN VoIP to DSCP EF
387
Example 12-58 Catalyst 6500 CatOS: Conditionally Trusted IP Phone + PC: Basic Model Example (Continued)
aggregate VVLAN-SIGNALING udp 10.1.110.0 0.0.0.255 any range 2000 2002
! Binds ACL to policer marks in-profile VVLAN Call-Signaling to DSCP CS3
aggregate VVLAN-ANY 10.1.110.0 0.0.0.255
! Binds ACL to policer and marks all other VVLAN traffic to DSCP 0
! Binds ACL to policer and marks in-profile PC Data traffic to DSCP 0
CAT6500-PFC2-CATOS> (enable) commit qos acl IPPHONE-PC-BASIC
CAT6500-PFC2-CATOS> (enable) set port qos 3/1 trust-device ciscoipphone
! Conditional trust (for Cisco IP Phones only)
CAT6500-PFC2-CATOS> (enable) set qos acl map IPPHONE-PC-BASIC 3/1
NOTE
show qos status

show qos maps
show port qos
show qos acl
show qos policer
show qos statistics
As previously mentioned, on nonGigabit Ethernet line cards that use 2Q2T transmit
queuing and 1Q4T receive queuing, a hardware limitation prevents the proper functioning
of port-based trust (which affects trust-cos, trust-ipprec, and trust-dscp). On such line cards,
a workaround ACL can be used to achieve trust functionality. For such an example, see the
section Catalyst 6500 CatOS QoS Verication Command: show port qos (see Example
12-50), earlier in this chapter.

The Catalyst 6500 Conditionally Trusted IP Phone: Advanced model leverages the dualrate policing capabilities of the PFC2/PFC3. In Example 12-59, the dual-rate policing
feature is applied to client-to-server ows to complement the Untrusted Server model.
Dual-rate policing, in this context, allows for graduated markdown of Interactive-Video
(from PCs), Transactional Data, and Bulk Data. Specically, in this example, Interactive-
388
Video is marked down to AF42 if it is in excess of 300 kbps but less than 500 kbps; if it is
greater than 500 kbps, it is marked down to Scavenger (CS1). Similarly, Transactional Data
and Bulk Data are marked down to AF22 and AF12 (respectively) if they are in excess of
3 Mbps but less than 5 Mbps; if they are in excess of 5 Mbps, they are both marked down
to Scavenger (CS1). All other policers are consistent with the single-rate policer model.
The Catalyst 6500 PFC2/PFC3 Conditionally Trusted Endpoint Dual-Rate Policing: IP
Phone + PC Advanced model is illustrated in Figure 12-25.
Figure 12-25 Catalyst 6500 PFC2/PFC3 Conditionally Trusted Endpoint Dual-Rate Policing: IP Phone + PC
(Advanced Model) Example
Start
VVLAN +
DSCP EF
Yes
128 kbps
Yes
Trust and Transmit
No
No
VVLAN +
DSCP AF31 or
CS3
Drop
Yes
Yes
No
No
VVLAN
ANY
32 kbps
Yes
Re-Mark to DSCP CS1

and Transmit
Yes
32 kbps
No
No
Re-Mark to DSCP CS3

and Transmit
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
DVLAN +
UDP 1638432767
Yes
Yes

and Transmit
Yes

and Transmit
300 kbps
No
No
500 kbps
Re-Mark to DSCP CS1

and Transmit
No
DVLAN +
SAP ACLs
Mission-Critical
Data ACLs
Yes
5 Mbps
No
No
DVLAN +
Lotus Notes ACLs
Transactional
Data ACLs
Yes
Yes
Re-Mark to DSCP CS1

and Transmit
Yes

and Transmit
Yes

and Transmit
3 Mbps
No
No
Re-Mark to DSCP 25
and Transmit
5 Mbps
Re-Mark to DSCP CS1
and Transmit
No
DVLAN +
IMAP ACLs
Bulk Data ACLs
Yes
Yes

and Transmit
Yes

and Transmit
3 Mbps
No
No
5 Mbps
Re-Mark to DSCP CS1
and Transmit
No
DVLAN
ANY
Yes
Yes
5 Mbps
No
Re-Mark to DSCP 0
and Transmit
Re-Mark to DSCP CS1
and Transmit
NOTE
389
The discrete trafc watermarks at which graduated markdown should occur are at the
network administrators discretion and vary among enterprises and applications.
Example 12-59 shows a conguration for a Catalyst 6500 CatOS Conditionally Trusted IP
Phone + PC: Advanced model.
Example 12-59 Catalyst 6500 CatOS: Conditionally Trusted IP Phone + PC: Advanced
Model Example
46 48 56
CAT6500-PFC2-CATOS> (enable) set qos cos-dscp-map 0 8 16 24 32 56
! Modifies default CoS-DSCP mapping so that CoS 5 is mapped to DSCP EF
CAT6500-PFC2-CATOS> (enable) set qos policed-dscp-map normal-rate 0,24,25:8
! Excess Data, Call-Signaling and MC-Data traffic is marked down to CS1
! Excess Bulk traffic is marked down from DSCP AF11 to AF12
! Violating Bulk traffic is marked down to DSCP CS1
! Excess Transactional Data traffic is marked down from AF21 to AF22
! Violating Transactional Data traffic is marked down to DSCP CS1
! Excess Interactive-Video traffic is marked down from AF41 to AF42
! Violating Interactive-Video traffic is marked down to DSCP CS1
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate VVLAN-VOICE
! Defines the policer for IP Phone VoIP traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate VVLAN-SIGNALING
! Defines the policer for IP Phone Call-Signaling traffic
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate VVLAN-ANY
! Defines the policer for any other traffic sourced from the VVLAN
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate PC-VIDEO
! Defines the Dual-Rate policer for Interactive-Video
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate MISSION-CRITICAL
! Defines the policer for Mission-Critical Data
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate TRANSACTIONAL
! Defines the Dual-Rate policer for Transactional Data
CAT6500-PFC2-CATOS> (enable) set qos policer aggregate BULK
! Defines the Dual-Rate policer for Bulk Data
continues
390
Example 12-59 Catalyst 6500 CatOS: Conditionally Trusted IP Phone + PC: Advanced
Model Example (Continued)

! Defines the policer for all other PC Data traffic
CAT6500-PFC2-CATOS> (enable) set qos acl ip IPPHONE-PC-ADVANCED dscp 46
aggregate VVLAN-VOICE udp 10.1.110.0 0.0.0.255 any range 16384 32767
! Binds ACL to policer and marks in-profile VVLAN VoIP to DSCP EF
aggregate VVLAN-SIGNALING tcp 10.1.110.0 0.0.0.255 any range 2000 2002
! Binds ACL to policer marks in-profile VVLAN Call-Signaling to DSCP CS3
aggregate VVLAN-ANY 10.1.110.0 0.0.0.255
! Binds ACL to policer and marks all other VVLAN traffic to DSCP 0
aggregate PC-VIDEO udp any any range 16384 32767
! Binds ACL to Dual-Rate policer and marks in-profile PC Video to AF41
aggregate MISSION-CRITICAL tcp any any range 3200 3203
aggregate MISSION-CRITICAL tcp any any eq 3600
aggregate TRANSACTIONAL tcp any any eq 1352
! Binds ACL to Dual-Rate policer and marks in-profile Lotus to AF21
aggregate BULK tcp any any eq 143
! Binds ACL to Dual-Rate policer and marks in-profile IMAP to AF11
aggregate BULK tcp any any eq 220
! Binds ACL to Dual-Rate policer and marks in-profile IMAP to AF11
! Binds ACL to policer and marks other in-profile PC data to DSCP 0
CAT6500-PFC2-CATOS> (enable) commit qos acl IPPHONE-PC-ADVANCED
CAT6500-PFC2-CATOS> (enable) set port qos 3/1 trust-device ciscoipphone
! Conditional trust (for Cisco IP Phones only)
CAT6500-PFC2-CATOS> (enable) set qos acl map IPPHONE-PC-ADVANCED 3/1
391
NOTE
show qos status

show qos maps
show port qos
show qos acl
show qos policer
show qos statistics
As previously mentioned, on nonGigabit Ethernet line cards that use 2Q2T transmit
queuing and 1Q4T receive queuing, a hardware limitation prevents the proper functioning
of port-based trust (which affects trust-cos, trust-ipprec, and trust-dscp). On such line cards,
a workaround ACL can be used to achieve trust functionality. For such an example, see the
section Catalyst 6500 CatOS QoS Verication Command: show port qos (see Example
12-50), earlier in this chapter.
Catalyst 6500: Queuing and Dropping

Although the Catalyst 6500 PFC performs classication, marking, mapping, and policing
functions, all queuing and dropping policies are administered by the Catalyst 6500 line
cards. This inevitably leads to per-line card hardware-specic capabilities and syntax when
it comes to conguring queuing and dropping.
As previously discussed in relation to other platforms that support ingress queuing, receive
queues are extremely difcult to congest, even in controlled lab environments. This is
especially so if access-edge policies, as detailed in this chapter, are used on all access-layer
switches.
Ingress congestion implies that the combined ingress rates of trafc exceed the switchs
processing capability and, thus, packets would need to be queued simply to gain access to
the switching fabric. On newer platforms, such as the Catalyst 6500 Supervisor 720, this
means that a combined ingress rate of more than 720 Gbps would have to be sent to the
switch, which is extremely unlikely.
To obviate such an extreme event, the Catalyst 6500 schedules ingress trafc through the
receive queues based on CoS values. In the default conguration, the scheduler assigns all
trafc with CoS 5 to the strict-priority queue (if present); in the absence of a strict-priority
queue, the scheduler assigns all trafc to the standard queues. All other trafc is assigned
to the standard queue(s) (with higher CoS values being assigned preference over lower CoS
values, wherever supported). Additionally, if a port is congured to trust CoS, the ingress
scheduler implements CoS value-based receive-queue drop thresholds, to avoid congestion
392
in received trafc. Thus, even if the extremely unlikely event of ingress congestion occurs,
the default settings for the Catalyst 6500 line card receive queues are more than adequate
to protect VoIP and Network-Control trafc.
Therefore, the focus of this section is on Catalyst 6500 egress/transmit queuing design
recommendations. At the time of writing, there are six main transmit queuing/dropping
options for Catalyst 6500 line cards:
2Q2TIndicates two standard queues, each with two congurable tail-drop

thresholds.
1P2Q1TIndicates one strict-priority queue and two standard queues, each with
one congurable WRED drop threshold. (However, each standard queue also has one
noncongurable tail-drop threshold.)
1P2Q2TIndicates one strict-priority queue and two standard queues, each with two
congurable WRED drop thresholds.
1P3Q1TIndicates one strict-priority queue and three standard queues, each with
one congurable WRED drop threshold. (However, each standard queue also has one
noncongurable tail-drop threshold.)
1P3Q8TIndicates one strict-priority queue and three standard queues, each with
eight congurable WRED drop thresholds. (However, each standard queue also has
one noncongurable tail-drop threshold.)
1P7Q8TIndicates one strict-priority queue and seven standard queues, each with
eight congurable WRED drop thresholds. (On 1p7q8t ports, each standard queue
also has one noncongurable tail-drop threshold.)
Almost all Catalyst 6500 line cards support a strict-priority queue, and, when supported,
the switch services trafc in the strict-priority transmit queue before servicing the standard
queues. When the switch is servicing a standard queue, after transmitting a packet, it checks
for trafc in the strict-priority queue. If the switch detects trafc in the strict-priority queue,
it suspends its service of the standard queue and completes service of all trafc in the strictpriority queue before returning to the standard queue.
Additionally, Catalyst 6500 line cards implement CoS value-based transmit-queue drop
thresholds to avoid congestion in transmitted trafc. WRED thresholds also can be dened
on certain line cards, where the CoS value of the packet (not the IP Precedence value,
although they likely will match) determines the WRED weight. WRED parameters include
a lower and upper threshold: The low WRED threshold is the queue level where (assigned)
trafc begins to be dropped selectively, and the high WRED threshold is the queue level
above which all (assigned) trafc is dropped. Furthermore, packets in the queue between
the low and high WRED thresholds have an increasing chance of being dropped as the
queue lls.
The transmit queuing and dropping capabilities can be returned with the following
commands.
393
CatOS:
show port capabilities

show port qos
show qos info
Cisco IOS:
show queueing interface
Table 12-2 includes the Catalyst 6500 line cards that were available at the time of this
writing, along with their respective queuing and dropping structures.
Table 12-2
Catalyst 6500 Line Card Queuing Structures
C2 (xCEF720)
Modules
Description
ReceiveQueue
Structure
Transmit Queue
Structure
Buffer Size
WS-X6704-10GE
Catalyst 6500 4-port

10 Gigabit Ethernet
module
1Q8T (8Q8T
with DFC3a)
1P7Q8T
16 MB per port
WS-X6724-SFP

Gigabit Ethernet SFP
module
1Q8T (2Q8T
with DFC3a)
1P3Q8T
1 MB per port
WS-X6748-GE-TX

10/100/1000 RJ-45
module
1Q8T (2Q8T
with DFC3a)
1P3Q8T
1 MB per port
WS-X6748-SFP

Gigabit Ethernet SFP
module
1Q8T (2Q8T
with DFC3a)
1P3Q8T
1 MB per port
ReceiveQueue
Structure
Transmit Queue
Structure
Buffer Size
Classic/CEF256
Ethernet Modules
Description
WS-X6024-10FL-MT

10BASE-FL MT-RJ
module
1Q4T
2Q2T
64 KB per port
WS-X6148-RJ21

10/100 RJ-21 module
(upgradeable to Voice)
1Q4T
2Q2T
128 KB per port
WS-X6148-RJ21V

10/100 Inline Power RJ-21
module
1Q4T
2Q2T
128 KB per port
WS-X6148-RJ45

10/100; RJ-45 module
1Q4T
2Q2T
128 KB per port
continues
394
Table 12-2
Catalyst 6500 Line Card Queuing Structures (Continued)
Classic/CEF256
Ethernet Modules
Description
ReceiveQueue
Structure
Transmit Queue
Structure
Buffer Size
WS-X6148-RJ45V
Catalyst 6500 48-port 10/

100 Inline Power RJ-45
module
1Q4T
2Q2T
128 KB per port
WS-X6148-GE-TX

10/100/1000 RJ-45
module
1Q2T
1P2Q2T
1 MB per 8 ports
WS-X6148V-GE-TX

10/100/1000 Inline Power
RJ-45 module
1Q2T
1P2Q2T
1 MB per 8 ports
WS-X6316-GE-TX

1000TX Gigabit Ethernet
RJ-45 module
1P1Q4T
1P2Q2T
512 KB per port
WS-X6324-100FX-MM Catalyst 6000 24-port

100FX MT-RJ MMF
module (with Enhanced
QoS)
1Q4T
2Q2T
128 KB per port
WS-X6324-100FX-SM

100FX MT-RJ SMF
module (with Enhanced
QoS)
1Q4T
2Q2T
128 KB per port
WS-X6348-RJ-21

10/100 RJ-21 module
1Q4T
2Q2T
128 KB per port
WS-X6348-RJ21V

module
1Q4T
2Q2T
128 KB per port
WS-X6348-RJ-45

10/100 RJ-45 module
1Q4T
2Q2T
128 KB per port
WS-X6348-RJ45V

module
1Q4T
2Q2T
128 KB per port
WS-X6408A-GBIC

Gigabit Ethernet module
(with Enhanced QoS;
requires GBICs)
1P1Q4T
1P2Q2T
512 KB per port
Table 12-2
395
Classic/CEF256
Ethernet Modules
Description
ReceiveQueue
Structure
Transmit Queue
Structure
Buffer Size
WS-X6416-GBIC

(requires GBICs)
1P1Q4T
1P2Q2T
512 KB per port
WS-X6416-GE-MT

Gigabit Ethernet MT-RJ
module
1P1Q4T
1P2Q2T
512 KB per port
WS-X6501-10GEX4
1-port 10 Gigabit Ethernet

module
1P1Q8T
1P2Q1T
64 MB per port
WS-X6502-10GE
Catalyst 6500 10 Gigabit

Ethernet Base module
(requires OIM)
1P1Q8T
1P2Q1T
64 MB per port
WS-X6516A-GBIC

(fabric enabled; requires
GBICs)
1P1Q4T
1P2Q2T
1 MB per port
WS-X6516-GBIC

GBICs)
1P1Q4T
1P2Q2T
512 KB per port
WS-X6516-GE-TX

Gigabit Ethernet Copper
module; (crossbar enabled)
1P1Q4T
1P2Q2T
512 KB per port
WS-X6524-100FX-MM Catalyst 6500 24-port

100FX MT-RJ module
(fabric enabled)
1P1Q0T
1P3Q1T
1 MB per port
WS-X6548-RJ-21

10/100 RJ-21 module
(fabric enabled)
1P1Q0T
1P3Q1T
1 MB per port
WS-X6548-RJ-45

10/100 RJ-45 module
(crossbar enabled)
1P1Q0T
1P3Q1T
1 MB per port
WS-X6548V-GE-TX

10/100/1000 Inline Power
RJ-45 module (fabric
enabled)
1Q2T
1P2Q2T
1 MB per 8 ports
continues
396
Table 12-2
Classic/CEF256
Ethernet Modules
Description
ReceiveQueue
Structure
Transmit Queue
Structure
Buffer Size
WS-X6548-GE-TX

10/100/1000 RJ-45
module (fabric enabled)
1Q2T
1P2Q2T
1 MB per 8 ports
WS-X6816-GBIC

GBICs)
1P1Q4T
1P2Q2T
512 KB per port
Design recommendations for each of these six main Catalyst 6500 queuing structures
follow.
Catalyst 6500: 2Q2T Queuing and Dropping

Line cards that support only 2Q2T queuing models have no provision for priority queuing.
Nonetheless, tuning the weighted round-robin (WRR) weights and the queue sizes can help
offset this limitation.
For example, if Q1 is to service Scavenger/Bulk Data (CoS 1) and Best-Effort (CoS 0)
trafc, assigning 30 percent of the buffer space to the rst queue is adequate; the
remaining 70 percent can be assigned to Q2.
The WRR weights can be set to the same ratio of 30:70 for servicing Q1:Q2.
Because the 2Q2T model supports congurable tail-drop thresholds, these can be tuned to
provide an additional layer of QoS granularity. For example, the rst queues rst threshold
can be set at 40 percent, to prevent Scavenger/Bulk Data trafc from dominating Q1.
Similarly, the second queues rst threshold can be set to 80 percent, to always allow some
room in the queue for VoIP. The second threshold of each queue always should be set to the
tail of the queue (100 percent).
After the queues and thresholds have been dened as such, CoS 1 (Scavenger/Bulk Data)
can be assigned to Q1T1; CoS 0 (Best Effort) can be assigned to Q1T2; CoS 2 (NetworkManagement and Transactional Data), CoS 3 (Call-Signaling and Mission-Critical Data),
CoS 4 (Interactive- and Streaming-Video), and CoS 6 and 7 (Internetwork and Network
Control) can be assigned to Q21T; and CoS 5 (VoIP) can be assigned to Q2T2.
Figure 12-26 illustrates these 2Q2T queuing recommendations.
397
Figure 12-26 Catalyst 6500 2Q2T Queuing Model

Application
DSCP
CoS
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
Interactive-Video
AF41
CoS 4
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
2Q2T
Q2T2
CoS 5
Q2T1
CoS 7
CoS 6
CoS 4
Queue 2
(70%)
CoS 3
CoS 2
Scavenger
CS1
CoS 1
Best-Effort
Q1T2
CoS 0
CoS 1
Queue 1
(30%)
Q1T1
Example 12-60 shows the Catalyst 6500 CatOS congurations to congure 2Q2T queuing
recommendations.
Example 12-60 Catalyst 6500 CatOS: 2Q2T Queuing Example
CAT6500-PFC2-CATOS> (enable) set qos txq-ratio 2q2t 30 70
! Sets the buffer allocations to 30% for Q1 and 70% for Q2
CAT6500-PFC2-CATOS> (enable) set qos wrr 2q2t 30 70
! Sets the WRR weights for 30:70 (Q1:Q2) bandwidth servicing
CAT6500-PFC2-CATOS> (enable) set qos drop-threshold 2q2t tx queue 1 40
5 100
100
! Sets Q1T1 to 40% to limit Scavenger/Bulk from dominating Q1
CAT6500-PFC2-CATOS> (enable) set qos drop-threshold 2q2t tx queue 2 80 100
! Sets Q2T1 to 80% to always have room in Q2 for VoIP
CAT6500-PFC2-CATOS> (enable) set qos map 2q2t tx 1 1 cos 1
! Assigns Scavenger/Bulk to Q1T1
! Assigns Best Effort to Q1T2
CAT6500-PFC2-CATOS> (enable) set qos map 2q2t tx 2 1 cos 2,3,4,6,7
! Assigns CoS 2,3,4,6 and 7 to Q2T1
! Assigns VoIP to Q2T2
398
show qos info cong 2q2t tx

show qos info runtime
show qos statistics
Catalyst 6500 CatOS QoS Verication Command: show qos info

cong 2q2t tx
The Catalyst 6500 CatOS show qos info cong 2q2t tx verication command displays the
queuing and dropping parameters for 2Q2T line cards.
In Example 12-61, CoS 1 is assigned to Q1T1; CoS 0 is assigned to Q1T2; CoS values 2,
3, 4, 6, and 7 are assigned to Q2T1; and CoS 5 is assigned to Q2T2. The rst thresholds are
set to 40 percent and 80 percent of their respective queues, and the second thresholds are
set to the tail of the queue. The size ratio has been allocated 30 percent for Q1 and 70
percent for Q2, and the WRR weights are set to 30:70 to service Q1 and Q2, respectively.
Example 12-61 show qos info cong 2q2t tx Verication for a Catalyst 6500-CatOS Switch
CAT6500-PFC2-CATOS> (enable) show qos info config 2q2t tx
QoS setting in NVRAM for 2q2t transmit:
QoS is enabled
Queue and Threshold Mapping for 2q2t (tx):
Queue Threshold CoS
----- --------- --------------1
1
1
1
2
0
2
1
2 3 4 6 7
2
2
5
Tx drop thresholds:
Queue # Thresholds - percentage
------- ------------------------------------1
5% 100%
40%
2
80% 100%
Tx WRED thresholds:
WRED feature is not supported for this port type.
Tx queue size ratio:
Queue # Sizes - percentage
------- ------------------------------------1
30%
2
70%
Tx WRR Configuration of ports with 2q2t:
Queue # Ratios
------- ------------------------------------1
30
2
70
399
Catalyst 6500 CatOS QoS Verication Command: show qos info runtime
The Catalyst 6500 CatOS show qos info runtime verication command reports similar
information as the show qos info cong command, but it displays the runtime information
(committed to the PFC and line card) instead of only the congured information.
In Example 12-62, CoS 1 is assigned to Q1T1; CoS 0 is assigned to Q1T2; CoS values 2,
3, 4, 6, and 7 are assigned to Q2T1; and CoS 5 is assigned to Q2T2. The rst thresholds are
set to 40 percent and 80 percent of their respective queues, and the second thresholds are
set to the tail of the queue. The size ratio has been allocated 30 percent for Q1 and 70
percent for Q2, and the WRR weights are set to 30:70 to service Q1 and Q2, respectively.
Example 12-62 show qos info runtime Verication for a Catalyst 6500-CatOS Switch
CAT6500-PFC3-CATOS> (enable) show qos info runtime 3/1
Run time setting of QoS:
QoS is enabled
Policy Source of port 3/1: Local
Tx port type of port 3/1 : 2q2t
Rx port type of port 3/1 : 1q4t
Interface type: port-based
ACL attached:
The qos trust type is set to untrusted.
Default CoS = 0
Queue and Threshold Mapping for 2q2t (tx):
Queue Threshold CoS
----- --------- --------------1
1
1
1
2
0
2
1
2 3 4 6 7
2
2
5
Queue and Threshold Mapping for 1q4t (rx):
All packets are mapped to a single queue.
Rx drop thresholds:
Rx drop thresholds are disabled.
Tx drop thresholds:
Queue # Thresholds - percentage (* abs values)
------- ------------------------------------1
5% (768
bytes)
100%
(15360
bytes)
40%
(6144
bytes)
100%
(15360
bytes)
2
80% (28672 bytes) 100% (35840 bytes)
Rx WRED thresholds:
Rx WRED feature is not supported for this port type.
Tx WRED thresholds:
WRED feature is not supported for this port type.
Tx queue size ratio:
Queue # Sizes - percentage (* abs values)
------- ------------------------------------1
30% (17408 bytes)
2
70% (37888 bytes)
Rx queue size ratio:
Rx queue size-ratio feature is not supported for this port type.
Tx WRR Configuration of ports with speed 10Mbps:
continues
400
Example 12-62 show qos info runtime Verication for a Catalyst 6500-CatOS Switch (Continued)
Queue # Ratios (* abs values)
------- ------------------------------------1
30 (7648 bytes)
2
70 (17840 bytes)
(*) Runtime information may differ from user configured setting due to hardware
granularity.
Example 12-63 shows the Catalyst 6500 IOS congurations to congure 2Q2T queuing
recommendations.
Example 12-63 Catalyst 6500 IOS: 2Q2T Queuing Example
CAT6500-PFC3-IOS(config)# interface range FastEthernet6/1 - 48
CAT6500-PFC3-IOS(config-if)# wrr-queue queue-limit 30 70
CAT6500-PFC3-IOS(config-if)# wrr-queue bandwidth 30 70
CAT6500-PFC3-IOS(config-if)#
CAT6500-PFC3-IOS(config-if)# wrr-queue threshold 1 40
5 100
100
! Sets Q1T1 to 40% to limit Scavenger/Bulk from dominating Q1
CAT6500-PFC3-IOS(config-if)# wrr-queue threshold 2 80 100
! Sets Q2T1 to 80% to always have room in Q2 for VoIP
CAT6500-PFC3-IOS(config-if)# wrr-queue cos-map 1 1 1
! Assigns Scavenger/Bulk to Q1T1
CAT6500-PFC3-IOS(config-if)# wrr-queue cos-map 2 1 2 3 4 6 7
! Assigns CoS 2,3,4,6 and 7 to Q2T1
! Assigns VoIP to Q2T2
CAT6500-PFC3-IOS(config-if)#end
CAT6500-PFC3-IOS#
Catalyst 6500 MLS QoS verication command:
Catalyst 6500 IOS QoS Verication Command: show queueing interface

The Catalyst 6500 IOS show queueing interface verication command displays the
queuing parameters for a given interface (according to the line cards capabilities).
In Example 12-64, the line card has 2Q2T transmit queuing. The WRR scheduling weights
are set to 30:70 to service Q1 and Q2, respectively. The transmit queue size ratios have been
allocated 30 percent for Q1 and 70 percent for Q2. The rst queues tail-drop thresholds are
set to 40 percent and 100 percent, while the second queues tail-drop thresholds are set to
401
80 percent and 100 percent. CoS 1 is assigned to Q1T1; CoS 0 is assigned to Q1T2; CoS
values 2, 3, 4, 6, and 7 are assigned to Q2T1; and CoS 5 is assigned to Q2T2.
Example 12-64 show queueing interface Verication for a Catalyst 6500 IOS Switch
CAT6500-PFC3-IOS#show queueing interface FastEthernet6/1
Interface FastEthernet6/1 queueing strategy: Weighted Round-Robin
Port QoS is enabled
Port is untrusted
Extend trust state: not trusted [COS = 0]
Default COS is 0
Queueing Mode In Tx direction: mode-cos
Transmit queues [type = 2q2t]:
Queue Id
Scheduling Num of thresholds
----------------------------------------1
WRR low
2
2
WRR high
2
WRR bandwidth ratios:
30[queue 1] 70[queue 2]
queue-limit ratios:
30[queue 1] 70[queue 2]
queue tail-drop-thresholds
-------------------------5[1] 100[2]
1
40[1]
100[2]
2
80[1] 100[2]
queue thresh cos-map
--------------------------------------1
1
1
1
2
0
2
1
2 3 4 6 7
2
2
5
<output truncated>
CAT6500-PFC3-IOS#
Catalyst 6500: 1P2Q1T Queuing and Dropping

The 1P2Q1T queuing model builds on the previous 2Q2T model, bringing with it the
advantages of strict-priority queuing (for VoIP) and a tunable WRED (not tail drop)
threshold per queue.
The term 1P2Q1T is a bit of a misnomer in the CatOS version of this queuing structure
because, in CatOS, there are actually two thresholds per queue: the tunable WRED
threshold and the noncongurable tail-of-the-queue (100 percent) tail-drop threshold.
Under such a model, buffer space can be allocated as follows: 30 percent for Scavenger/
Bulk Data plus Best-Effort queue (Q1), 40 percent for Q2, and 30 percent for the PQ (Q3).
The WRR weights for Q1 and Q2 (for dividing the remaining bandwidth, after the priority
queue has been serviced fully) can be set to 30:70, respectively, for Q1:Q2.
Under the 1P2Q1T model, each queues WRED threshold is dened with a lower and upper
limit. For example, the WRED threshold 40:80 indicates that packets assigned to this
402
WRED threshold will begin being randomly dropped when the queue lls to 40 percent and
that these packets will be tail-dropped if the queue lls beyond 80 percent.
Furthermore, in CatOS within the 1P2Q1T queuing structure, each CoS value can be assigned
to a queue and a WRED threshold or just to a queue. When assigned to a queue (only), the
CoS value will be limited only by the tail of the queue. (In other words, it is assigned to
the queue with a tail drop threshold of 100 percent.)
Thus (in CatOS), the tunable WRED threshold for Q1 can be set to 40:80, meaning that
Scavenger/Bulk Data will be WRED-dropped if Q1 lls to 40 percent and will be taildropped if Q1 lls past 80 percent of capacity. This prevents Scavenger/Bulk Data from
drowning out Best-Effort trafc in Q1. The WRED threshold for Q2 can be set to 70:80 to
provide congestion avoidance for all applications assigned to it and to ensure that there will
always be room in the queue to service Network and Internetwork Control trafc.
Therefore, when the queues and thresholds have been dened as such, CoS 1 (Scavenger/
Bulk Data) can be assigned to Q1T1; CoS 0 (Best-Effort) can be assigned to Q1 only (tail);
CoS 2 (Network-Management and Transactional Data), CoS 3 (Call-Signaling and MissionCritical Data) and CoS 4 (Interactive- and Streaming-Video) can be assigned to Q2T1; CoS
6 and 7 (Internetwork and Network Control) can be assigned to Q2 only (tail); and CoS 5
(VoIP) can be assigned to Q3 (the PQ).
Figure 12-27 illustrates these 1P2Q1T queuing recommendations.
Figure 12-27 Catalyst 6500 1P2Q1T Queuing Model (CatOS Supports 1P2Q2T)
Application
DSCP
CoS
1P2Q1T
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
CoS 7
Interactive-Video
AF41
CoS 4
CoS 6
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
Scavenger
CS1
CoS 1
Best-Effort
CoS 5
Q4
Q3 (5%)
Priority Queue
Q2T2
Q2T1
CoS 4
Queue 2
(40%)
CoS 3
CoS 2
Q1T2
CoS 0
CoS 1
Queue 1
(30%)
Q1T1
403
Example 12-65 shows the Catalyst 6500 CatOS congurations to congure 1P2Q1T
queuing recommendations.
Example 12-65 Catalyst 6500 CatOS: 1P2Q1T (Technically, 1P2Q2T) Queuing Example
CAT6500-PFC2-CATOS> (enable) set qos txq-ratio 1p2q1t 30 40 30
! Sets the buffer allocations to 30% for Q1, 40% for Q2, 30% for Q3 (PQ)
CAT6500-PFC2-CATOS> (enable) set qos wrr 1p2q1t 30 70
CAT6500-PFC2-CATOS> (enable) set qos wred 1p2q1t tx queue 1 1:5
40:80
! Sets Q1 WRED Threshold to 40:80 to limit Scavenger/Bulk from dominating Q1
70:80
! Sets Q2 WRED Threshold to 70:80 to force room for Network Control traffic
CAT6500-PFC2-CATOS> (enable) set qos map 1p2q1t tx 1 1 cos 1
! Assigns Scavenger/Bulk to Q1 WRED Threshold
CAT6500-PFC2-CATOS> (enable) set qos map 1p2q1t tx 1 cos 0
! Assigns Best Effort to Q1 tail (100%) threshold
CAT6500-PFC2-CATOS> (enable) set qos map 1p2q1t tx 2 1 cos 2,3,4
! Assigns CoS 2,3,4 to Q2 WRED Threshold
CAT6500-PFC2-CATOS> (enable) set qos map 1p2q1t tx 2 cos 6,7
! Assigns Network/Internetwork Control to Q2 tail (100%) threshold
! Assigns VoIP to PQ (Q3)
NOTE
show qos info cong 1p2q1t tx

show qos statistics
The Catalyst 6500 CatOS show qos info verication commands are reasonably similar for
each queuing structure and, as such, are not detailed for each queuing model example.
In Cisco IOS, for any 1PxQyT queuing structure, setting the size of the priority queue is not
supported. The only exception to this rule is the 1P2Q2T structure, in which the priority
queue (Q3) indirectly is set to equal Q2s size. Therefore, in all examples of Catalyst 6500 IOS
queuing structure congurations that follow that followexcept for the 1P2Q2T
exampleonly the sizes of the standard queues are being set.
404
Furthermore, specic to the 1P2Q1T queuing structure, CoS values cannot be mapped to
the tail of the queue, as in CatOS. CoS values can be mapped only to the single WRED
threshold for each queue. Therefore, the 1P2Q1T queuing and dropping recommendation
requires some slight alterations for Cisco IOS. These include changing Q1T1s WRED
threshold to 80:100 and, likewise, changing Q2T1s WRED threshold to 80:100.
The syntax logic for setting WRED thresholds in Cisco IOS is different than in CatOS. In
CatOS, minimum and maximum WRED thresholds are set on the same line; in Cisco IOS,
minimum and maximum WRED thresholds are set on different lines.
After these WRED thresholds have been altered, CoS 1 (Scavenger/Bulk Data) and CoS 0
(Best-Effort) can be assigned to Q1T1; CoS 2 (Network-Management and Transactional
Data), CoS 3 (Call-Signaling and Mission-Critical Data), CoS 4 (Interactive- and StreamingVideo), and CoS 6 and 7 (Internetwork and Network Control) can be assigned to Q2T1; and
CoS 5 (VoIP) can be assigned to Q3 (the PQ).
Example 12-66 shows the Catalyst 6500 IOS congurations to congure 1P2Q1T queuing
recommendations.
Example 12-66 Catalyst 6500 IOS: 1P2Q1T Queuing Example
CAT6500-PFC3-IOS(config)#interface TenGigabitEthernet1/1
CAT6500-PFC3-IOS(config-if)# wrr-queue queue-limit 30 40
CAT6500-PFC3-IOS(config-if)# wrr-queue bandwidth 30 70
CAT6500-PFC3-IOS(config-if)# wrr-queue random-detect min-threshold 1
! Sets Min WRED Threshold for Q1T1 to 80%
CAT6500-PFC3-IOS(config-if)# wrr-queue random-detect max-threshold 1
! Sets Max WRED Threshold for Q1T1 to 100%
CAT6500-PFC3-IOS(config-if)# wrr-queue random-detect min-threshold 2
CAT6500-PFC3-IOS(config-if)# wrr-queue random-detect max-threshold 2
CAT6500-PFC3-IOS(config-if)# wrr-queue cos-map 1 1 1 0
! Assigns Scavenger/Bulk and Best Effort to Q1 WRED Threshold
! Assigns CoS 2,3,4,6 and 7 to Q2 WRED Threshold 1
CAT6500-PFC3-IOS(config-if)# priority-queue cos-map 1 5
1
80
100
80
100
405

The 1P2Q2T queuing model is essentially identical to the 1P2Q1T model, except that it
supports two congurable WRED thresholds per queue.
Under a 1P2Q2T model, buffer space can be allocated as follows: 40 percent for Q1 (the
Scavenger/Bulk Data + Best-Effort queue), 30 percent for Q2 (the preferential queue), and
30 percent for the Q3 (the priority queue).
The WRR weights for Q1 and Q2 (for dividing the remaining bandwidth, after the priority
queue has been serviced fully) remain at 30:70, respectively, for Q1:Q2.
Under the 1P2Q2T model, each WRED threshold is dened with a lower and upper limit.
Therefore, the rst WRED threshold for Q1 can be set to 40:80, so that Scavenger/Bulk
Data trafc can be WRED-dropped if Q1 hits 40 percent and can be tail-dropped if Q1
exceeds 80 percent of its capacity (this prevents Scavenger/Bulk Data from drowning out
Best-Effort trafc in Q1). The second WRED threshold for Q1 can be set to 80:100 to
provide congestion avoidance for Best-Effort trafc.
Similarly, the rst WRED threshold of Q2 can be set to 70:80, and the second can be set to
80:100. In this manner, congestion avoidance will be provided for all trafc types in Q2,
and there will always be room in the queue to service Network and Internetwork Control
trafc.
Therefore, after the queues have been dened as mentioned previously, CoS 1 (Scavenger/
Bulk Data) can be assigned to Q1T1; CoS 0 (Best-Effort) can be assigned to Q1T2; CoS 2
(Network-Management and Transactional Data), CoS 3 (Call-Signaling and MissionCritical Data), and CoS 4 (Interactive- and Streaming-Video) can be assigned to Q2T1; CoS
6 and 7 (Internetwork and Network Control) can be assigned to Q2T2; and CoS 5 (VoIP)
can be assigned to Q3T1 (the PQ).
406

Application
DSCP
CoS
1P2Q2T
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
CoS 7
Interactive-Video
AF41
CoS 4
CoS 6
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
Scavenger
CS1
CoS 1
Best-Effort
CoS 5
Q3 (30%)
Q4 (5%)
Priority Queue
Q2T2
Q2T1
CoS 4
Queue 2
(30%)
CoS 3
CoS 2
Q2T1
CoS 0
CoS 1
Queue 1
(40%)
Q1T1
Example 12-67 Catalyst 6500 CatOS: 1P2Q2T Queuing Example
CAT6500-PFC2-CATOS> (enable) set qos txq-ratio 1p2q2t 40 30 30
! Sets the buffer allocations to 40% for Q1, 30% for Q2, 30% for Q3 (PQ)
CAT6500-PFC2-CATOS> (enable) set qos wrr 1p2q2t 30 70
CAT6500-PFC2-CATOS> (enable) set qos wred 1p2q2t tx queue 1 40:80 80:100
! Sets Q1 WRED T1 to 40:80 to limit Scavenger/Bulk from dominating Q1
! Sets Q1 WRED T2 to 80:100 to provide congestion-avoidance for Best Effort
! Sets Q2 WRED T1 to 70:80 to provide congestion-avoidance
! Sets Q2 WRED T2 to 80:100 to force room for Network Control traffic
! Assigns Scavenger/Bulk to Q1 WRED Threshold 1
! Assigns Best Effort to Q1 WRED Threshold 2
! Assigns CoS 2,3,4 to Q2 WRED Threshold 1
CAT6500-PFC2-CATOS> (enable) set qos map 1p2q2t tx 2 2 cos 6,7
! Assigns Network/Internetwork Control to Q2 WRED Threshold 2
407
Example 12-67 Catalyst 6500 CatOS: 1P2Q2T Queuing Example (Continued)

! Assigns VoIP to PQ

show qos statistics
Example 12-68 shows the compatible Catalyst 6500 IOS congurations to congure
1P2Q1T queuing recommendations. Notice that the buffer allocation for the PQ (Q3) is not
congurable but, by default (for the 1P2Q2T queuing structure only), is set to equal the size
dened for Q2. Therefore, Q1 is set to 40 percent and Q2 is set to 30 percent, which
indirectly sets Q3 to match at 30 percent.
recommendations.
interface range GigabitEthernet4/1 - 8
CAT6500-PFC3-IOS(config)#interface
CAT6500-PFC3(config-if-range)# wrr-queue queue-limit 40 30
! Indirectly sets PQ (Q3) size to equal Q2 (which is set to 30%)
CAT6500-PFC3(config-if-range)# wrr-queue bandwidth 30 70
CAT6500-PFC3(config-if-range)#
CAT6500-PFC3(config-if-range)# wrr-queue random-detect min-threshold 1 40 80
! Sets Min WRED Thresholds for Q1T1 and Q1T2 to 40 and 80, respectively
CAT6500-PFC3(config-if-range)# wrr-queue random-detect max-threshold 1 80 100
! Sets Max WRED Thresholds for Q1T1 and Q1T2 to 80 and 100, respectively
CAT6500-PFC3(config-if-range)# wrr-queue random-detect min-threshold 2 70 80
CAT6500-PFC3(config-if-range)# wrr-queue random-detect max-threshold 2 80 100
CAT6500-PFC3(config-if-range)# wrr-queue cos-map 1 1 1
CAT6500-PFC3(config-if-range)# wrr-queue cos-map 2 1 2 3 4
! Assigns CoS 2,3,4 to Q2 WRED Threshold 1
CAT6500-PFC3(config-if-range)# wrr-queue cos-map 2 2 6 7
! Assigns Network/Internetwork Control to Q2 WRED Threshold 2
continues
408
Example 12-68 Catalyst 6500 IOS: 1P2Q2T Queuing Example (Continued)

CAT6500-PFC3(config-if-range)# priority-queue cos-map 1 5
! Assigns VoIP to PQ
CAT6500-PFC3(config-if-range)#end
CAT6500-PFC3-IOS#

The 1P3Q1T queuing structure is identical to the 1P2Q1T structure, except that an
additional standard queue has been added to it and that it does not support tuning the
transmit size ratios. Under this model, Q4 is the strict-priority queue.
The WRR weights for the standard queues (Q1, Q2, Q3), for dividing the remaining
bandwidth after the priority queue has been serviced serviced, can be set to 5:25:70,
respectively, for Q1:Q2:Q3.
In CatOS, within the 1P3T1T queuing structure, each CoS value can be assigned to a queue
and a WRED threshold or just to a queue. When assigned to a queue (only), the CoS value
is limited only by the tail of the queue (in other words, it is assigned to the queue with a tail
drop threshold of 100 percent). Therefore, CatOS essentially supports 1P3Q2T for this type
of line card.
Thus, the tunable WRED threshold for Q1 can be set to 80:100 to provide congestion avoidance for Scavenger/Bulk Data trafc. The WRED threshold for Q2 similarly can be set to
80:100 to provide congestion avoidance on all Best-Effort ows. The WRED threshold for
Q3 can be set to 70:80, to provide congestion avoidance for all applications assigned to it
and to ensure that there will always be room in the Q3 to service Network and Internetwork
Control trafc.
(Network-Management and Transactional Data), CoS 3 (Call-Signaling and Mission-Critical
Data), and CoS 4 (Interactive- and Streaming-Video) can be assigned to Q3T1; CoS 6 and
7 (Internetwork and Network Control) can be assigned to Q3 (tail); and CoS 5 (VoIP) can
be assigned to Q4 (the PQ).
409
Figure 12-29 Catalyst 6500 1P3Q1T Queuing Model (CatOS Supports 1P3Q2T)
Application
DSCP
CoS
1P3Q1T
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
CoS 7
Interactive-Video
AF41
CoS 4
CoS 6
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
Scavenger
CS1
CoS 1
Best-Effort
CoS 5
Q4
Q4 (5%)
Priority Queue
Q3T2
Q3T1
CoS 4
Queue 3
(70%)
CoS 3
CoS 2
CoS 0
CoS 1
Queue 2
(25%)
Queue 1 (5%)
Example 12-69 Catalyst 6500 CatOS: 1P3Q1T (Technically, 1P3Q2T) Queuing Example
CAT6500-PFC2-CATOS> (enable) set qos wrr 1p3q1t 5 25 70
! Sets the WRR weights for 5:25:70 (Q1:Q2:Q3) bandwidth servicing
! Sets Q1 WRED T1 to 80:100 to provide congestion-avoidance for Scavenger
1:100
! Sets Q2 WRED T1 to 80:100 to provide congestion-avoidance for Best Effort
1:80
! Sets Q3 WRED T1 to 70:80 to provide congestion-avoidance for CoS 2,3,4
! and to force room (via tail-drop) for Network Control traffic
! Assigns Scavenger/Bulk to Q1 WRED Threshold 1 (80:100)
! Assigns Best Effort to Q2 WRED Threshold 1 (80:100)
! Assigns CoS 2,3,4 to Q3 WRED Threshold 1 (70:80)
CAT6500-PFC2-CATOS> (enable) set qos map 1p3q1t tx 3 cos 6,7
! Assigns Network/Internetwork Control to Q3 Tail (100%)
continues
410
Example 12-69 Catalyst 6500 CatOS: 1P3Q1T (Technically, 1P3Q2T) Queuing Example (Continued)

show qos statistics
In Cisco IOS, the 1P3Q1T, 1P3Q8T, and 1P7Q8T queuing structures can be congured to
use tail drop or WRED. By default, WRED is disabled. Therefore, it is good practice to
always explicitly enable WRED on a queue before setting WRED thresholds for these
queuing structures.
Additionally, in Cisco IOS, the 1P3Q1T queuing structure does not support mapping CoS
values to the tail of the queue (only to the single WRED threshold). Therefore, the queuing
recommendation requires slight alterations for Cisco IOS: changing all three WRED
thresholds to 80:100 and mapping CoS values 2, 3, 4, 6, and 7 to Q3T1.
recommendations.
CAT6500-PFC3-IOS(config)# interface range FastEthernet3/1 - 48
CAT6500-PFC3-IOS(config-if)# wrr-queue bandwidth 5 25 70
CAT6500-PFC3(config-if-range)# wrr-queue random-detect 1
! Enables WRED on Q1
CAT6500-PFC3-IOS(config-if)# wrr-queue random-detect min-threshold 1 80
CAT6500-PFC3-IOS(config-if)# wrr-queue random-detect max-threshold 1 100
411

! Assigns Scavenger/Bulk to Q1 WRED Threshold 1 (80:100)
! Assigns Best Effort to Q2 WRED Threshold 1 (80:100)
! Assigns CoS 2,3,4,6 and 7 to Q3 WRED Threshold 1 (80:100)
CAT6500-PFC3-IOS#

The 1P3Q8T queuing structure is identical to the 1P3Q1T structure, except it has eight
tunable WRED thresholds per queue (instead of one) and it also supports tuning the
transmit size ratios. Under this model, Q4 is the strict-priority queue.
Under a 1P3Q8T model, buffer space can be allocated as follows: 5 percent for the Scavenger/
Bulk Data queue (Q1), 25 percent for the Best-Effort queue (Q2), 40 percent for the
preferential queue (Q3), and 30 percent for the strict-priority queue (Q4).
The WRR weights for the standard queues (Q1, Q2, Q3), for dividing the remaining bandwidth
after the priority queue has been serviced fully, can be set to 5:25:70, respectively, for
Q1:Q2:Q3.
The tunable WRED threshold for Q1 can be set to 80:100 to provide congestion avoidance
to Scavenger/Bulk Data trafc. The WRED threshold for Q2 similarly can be set to 80:100
to provide congestion avoidance on all Best-Effort ows.
The 1P3Q8T queuing structures support for up to eight WRED thresholds per queue allows
for additional QoS granularity for the applications sharing Q3. Because only ve discrete
CoS values are sharing this queue, only ve of eight thresholds need to be dened for
subqueue QoS. For example, Q3T1 could be set to 50:60, Q3T2 could be set to 60:70, Q3T3
could be set to 70:80, Q3T4 could be set to 80:90, and Q3T5 could be set to 90:100.
(Interactive- and Streaming-Video) can be assigned to Q3T1; CoS 2 (Network-Management
and Transactional Data) can be assigned to Q3T2; CoS 3 (Call-Signaling and Mission-Critical
412
Data) can be assigned to Q3T3; CoS 6 (Internetwork Control) can be assigned to Q3T4;
CoS 7 (Internetwork and Network Control) can be assigned to Q3T5; and CoS 5 (VoIP) can
be assigned to Q4 (the PQ).
Application
DSCP
CoS
Network Control
CoS 7
CS6
CoS 6
EF
CoS 5
1P3Q8T
CoS 5
Voice
Interactive-Video
AF41
Q4 (5%)
Priority Queue
CoS 7
Q3T5
CoS 6
Q3T4
CoS 3
Q3T3
CoS 2
Q3T2
CoS 4
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
Q3T1
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
Scavenger
CS1
CoS 1
Best-Effort
Queue 3
(40%)
CoS 4
Q2T1
CoS 0
CoS 1
Queue 2
(25%)
Queue 1 (5%)
Q1T1
Example 12-71 shows the Catalyst 6500 (PFC3) CatOS congurations to congure
1P3Q8T queuing recommendations.
Example 12-71 Catalyst 6500 (PFC3) CatOS: 1P3Q8T Queuing Example
CAT6500-PFC3-CATOS> (enable) set qos txq-ratio 1p3q8t 5 25 40 30
! Allocates 5% for Q1, 25% for Q2, 40% for Q3 and 30% for Q4 (PQ)
CAT6500-PFC3-CATOS> (enable) set qos wrr 1p3q8t 5 25 70
100:100 100:100 100:100 100:100 100:100 100:100
! Sets Q1 WRED T1 to 80:100 and all other Q1 WRED Thresholds to 100:100
100:100 100:100 100:100 100:100 100:100 100:100
CAT6500-PFC3-CATOS> (enable) set qos wred 1p3q8t tx queue 3 50:60 60:70 70:80
80:90 90:100 100:100 100:100 100:100
! Sets Q3 WRED T1 to 50:60, Q3T2 to 60:70, Q3T3 to 70:80,
413
Example 12-71 Catalyst 6500 (PFC3) CatOS: 1P3Q8T Queuing Example (Continued)
! Q3T4 to 80:90, Q3T5 to 90:100
! and the other two Q3 WRED Thresholds to 100:100
! Assigns Video to Q3 WRED Threshold 1
! Assigns Net-Mgmt and Transactional Data to Q3 WRED T2
! Assigns Call-Signaling and Mission-Critical Data to Q3 WRED T3
! Assigns Internetwork-Control (IP Routing) to Q3 WRED T4
! Assigns Network-Control (Spanning Tree) to Q3 WRED T5
! Assigns VoIP to the PQ (Q4)
Catalyst 6500 (PFC3) CatOS QoS verication commands:

show qos statistics
Example 12-72 shows the Catalyst 6500 (PFC3) IOS congurations to congure 1P3Q8T
CAT6500-PFC3-IOS(config)# interface range GigabitEthernet1/1 - 48
CAT6500-PFC3-IOS(config-if)# wrr-queue queue-limit 5 25 40
! Allocates 5% for Q1, 25% for Q2 and 40% for Q3
CAT6500-PFC3-IOS(config-if)# wrr-queue bandwidth 5 25 70
100 100 100 100 100 100 100
! Sets Min WRED Threshold for Q1T1 to 80% and all others to 100%
continues
414

100 100 100 100 100 100 100
! Sets Max WRED Threshold for Q1T1 to 100% and all others to 100%
100 100 100 100 100 100 100
100 100 100 100 100 100 100
60 70 80 90 100 100 100
! Sets Min WRED Threshold for Q3T1 to 50%, Q3T2 to 60%, Q3T3 to 70%
! Q3T4 to 80%, Q3T5 to 90% and all others to 100%
70 80 90 100 100 100 100
! Sets Max WRED Threshold for Q3T1 to 60%, Q3T2 to 70%, Q3T3 to 80%
! Q3T4 to 90%, Q3T5 to 100% and all others to 100%
CAT6500-PFC3-IOS#

The 1P7Q8T queuing structure adds four standard queues to the 1P3Q8T structure and
moves the PQ from Q4 to Q8. Otherwise, it is identical.
415
Under a 1P7Q8T model, buffer space can be allocated as follows: 5 percent for the
Scavenger/Bulk Data queue (Q1), 25 percent for the Best-Effort queue (Q2), 10 percent for
the Video queue (Q3), 10 percent for the Network-Management/Transactional Data queue
(Q4), 10 percent for the Call-Signaling/Mission-Critical Data queue (Q5), 5 percent for the
Internetwork Control queue (Q6), 5 percent for the Network Control queue (Q7), and 30
percent for the PQ (Q8).
The WRR weights for the standard queues (Q1 through Q7), for dividing the remaining
bandwidth, after the priority queue has been serviced fully, can be set to 5:25:20:20:20:5:5,
respectively, for Q1 through Q7.
Because eight queues are available, each CoS value can be assigned to its own exclusive
queue. WRED can be enabled on each queue to provide it with congestion avoidance, by
setting the rst WRED threshold of each queue to 80:100. All other WRED thresholds can
remain at 100:100.
Therefore, when the queues and thresholds have been dened as mentioned previously,
CoS 1 (Scavenger/Bulk Data) can be assigned to Q1T1; CoS 0 (Best-Effort) can be
assigned to Q2T1; CoS 4 (Interactive- and Streaming-Video) can be assigned to Q3T1;
CoS 2 (Network-Management and Transactional Data) can be assigned to Q4T1; CoS 3
(Call-Signaling and Mission-Critical Data) can be assigned to Q5T1; CoS 6 (Internetwork
Control) can be assigned to Q6T1; CoS 7 (Internetwork and Network Control) can be
assigned to Q7T1; and CoS 5 (VoIP) can be assigned to Q8 (the PQ).
Application
DSCP
CoS
Network Control
CoS 7
CS6
CoS 6
Voice
EF
CoS 5
Interactive-Video
AF41
CoS 4
Streaming-Video
CS4
CoS 4
DSCP 25
CoS 3
1P7Q8T
CoS 5
Q8 (PQ)
CoS 7
Q7 (5%)
Q7T1
CoS 6
Q6 (5%)
Q6T1
CoS 3
Q5 (20%)
CoS 2
Q4 (20%)
CoS 4
Q3 (20%)
CoS 0
Q2 (25%)
Q5T1
Call-Signaling
AF31/CS3
CoS 3
Transactional Data
AF21
CoS 2
Network-Management
CS2
CoS 2
Bulk Data
AF11
CoS 1
Scavenger
CS1
CoS 1
Best-Effort
Q4T1
Q3T1
Q2T1
CoS 1
Q1 (5%)
Q1T1
416
Example 12-73 shows the Catalyst 6500 (PFC3) CatOS congurations to congure
1P7Q8T queuing recommendations.
Example 12-73 Catalyst 6500 (PFC3) CatOS: 1P7Q8T Queuing Example
CAT6500-PFC3-CATOS> (enable) set qos txq-ratio 1p7q8t 5 25 10 10 10 5 5 30
! Allocates 5% to Q1, 25% to Q2, 10% to Q3, 10% to Q4,
! Allocates 10% to Q5, 5% to Q6, 5% to Q7 and 30% to the PQ (Q8)
CAT6500-PFC3-CATOS> (enable) set qos wrr 1p7q8t 5 25 20 20 20 5 5
! Sets the WRR weights for 5:25:20:20:20:5:5 (Q1 through Q7)
100:100 100:100 100:100 100:100 100:100 100:100
100:100 100:100 100:100 100:100 100:100 100:100
100:100 100:100 100:100 100:100 100:100 100:100
100:100 100:100 100:100 100:100 100:100 100:100
100:100 100:100 100:100 100:100 100:100 100:100
100:100 100:100 100:100 100:100 100:100 100:100
100:100 100:100 100:100 100:100 100:100 100:100
417
Catalyst 6500 (PFC3) CatOS QoS verication commands:

show qos statistics
Example 12-74 shows the Catalyst 6500 (PFC3) IOS congurations to congure 1P7Q8T
Example 12-74 Catalyst 6500 (PFC3) IOS: 1P7Q8T Queuing Example
interface range TenGigabitEthernet4/1 - 4
CAT6500-PFC3(config-if-range)# wrr-queue queue-limit 5 25 10 10 10 5 5
! Allocates 10% to Q5, 5% to Q6 and 5% to Q7
CAT6500-PFC3(config-if-range)# wrr-queue bandwidth 5 25 20 20 20 5 5
CAT6500-PFC3(config-if-range)# wrr-queue random-detect min-threshold 1 80
100 100 100 100 100 100 100
CAT6500-PFC3(config-if-range)# wrr-queue random-detect max-threshold 1 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
continues
418
Example 12-74 Catalyst 6500 (PFC3) IOS: 1P7Q8T Queuing Example (Continued)
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
CAT6500-PFC3(config-if-range)# priority-queue cos-map 1 5
CAT6500-PFC3-IOS#
419
Catalyst 6500: PFC3 Distribution-Layer (Cisco IOS) Per-User

Microow Policing
In general, superior defense strategies have multiple lines of defense. In the context of the
campus designs discussed in this chapter, there is a main line of defense against DoS/worm
attack trafc at the access-layer edges. This line of defense can be bolstered at the distribution
layer whenever Catalyst 6500 Sup720s (PFC3s) are deployed there. This can be done by
leveraging the PFC3 feature of per-user microow policing.
In Example 12-75, trafc has been assumed to be correctly classied. This might or might
not be a valid assumption. If it is suspected to be invalid, ACLs should be used to identify
the ows (instead of using DSCP markings). In either case, various ow types can be
ltered as they arrive at the distribution layer to see if they conform to the normal limits that
have been set for the enterprise. Each ow is examined by source IP address; if a source is
transmitting out of prole, the excess trafc can be dropped or marked down. In this
manner, spurious ows can be contained, even if access-layer switches (such as the Catalyst
2950, discussed earlier in this chapter) do not support granular policing or if policing has
been miscongured on an access-layer switch.
In this manner, the distribution-layer Catalyst 6500 PFC3 can catch any DoS/worm attack
ows that might have slipped through the access-layer net.
Example 12-75 Catalyst 6500 (PFC3) IOSDistribution-Layer Per-User Microow Policing Example
CAT6500-PFC3-IOS(config)#mls
mls qos map policed-dscp normal 0 24 26 34 36 to 8
! Excess traffic marked 0,CS3,AF31,AF41 or AF42 will be remarked to CS1
CAT6500-PFC3-IOS(config)#
CAT6500-PFC3-IOS(config)#class-map match-all VOIP
CAT6500-PFC3-IOS(config-cmap)# match ip dscp ef
CAT6500-PFC3-IOS(config-cmap)#class-map match-all INTERACTIVE-VIDEO
CAT6500-PFC3-IOS(config-cmap)# match ip dscp af41 af42
CAT6500-PFC3-IOS(config-cmap)#class-map match-all CALL-SIGNALING
CAT6500-PFC3-IOS(config-cmap)# match ip dscp cs3 af31
CAT6500-PFC3-IOS(config-cmap)#class-map match-all BEST-EFFORT
CAT6500-PFC3-IOS(config-cmap)# match ip dscp 0
CAT6500-PFC3-IOS(config-cmap)#
CAT6500-PFC3-IOS(config-cmap)#
CAT6500-PFC3-IOS(config-cmap)#policy-map
PER-USER-POLICING
conform-action
transmit exceed-action
CAT6500-PFC3-IOS(config-pmap)# class VOIP
CAT6500-PFC3-I(config-pmap-c)# police flow mask src-only 128000 8000
conform-action transmit exceed-action drop
! No source can send more than 128k worth of DSCP EF traffic
CAT6500-PFC3-I(config-pmap-c)# class INTERACTIVE-VIDEO
continues
420
Example 12-75 Catalyst 6500 (PFC3) IOSDistribution-Layer Per-User Microow Policing Example (Continued)
conform-action transmit exceed-action policed-dscp-transmit
! Excess IP/VC traffic from any source is marked down to CS1
CAT6500-PFC3-I(config-pmap-c)# class CALL-SIGNALING
! Excess Call-Signaling traffic from any source is marked down to CS1
CAT6500-PFC3-I(config-pmap-c)# class BEST-EFFORT
! Excess PC Data traffic from any source is marked down to CS1
CAT6500-PFC3-I(config-pmap-c)# exit
CAT6500-PFC3-IOS(config-pmap)#exit
interface range GigabitEthernet4/1 - 4
CAT6500-PFC3(config-if-range)# mls qos trust dscp
CAT6500-PFC3(config-if-range)# service-policy input PER-USER-POLICING
! Attaches Per-User Microflow policing policy to Uplinks from Access
CAT6500-PFC3-IOS#
Catalyst 6500 MLS QoS verication commands:
show mls qos

show class-map
show policy-map
WAN Aggregator/Branch Router Handoff

Considerations
A nal consideration in campus Qos design is the campus-to-WAN (or VPN) handoff. In
the case of a branch, this equates to a handoff from the branch switch to the branch router.
In either case, a major speed mismatch is impending because Gigabit Ethernet/Fast
Ethernet campus networks are connecting to WAN links that might be only a few megabits
(if that).
Granted, the WAN aggregation routers and branch routers have advanced QoS mechanisms
to prioritize trafc on their links, but it is critical to keep in mind that Cisco router QoS is
performed in Cisco IOS software, while Catalyst switch QoS is performed in ASIC
hardware.
WAN Aggregator/Branch Router Handoff Considerations
421
Therefore, the optimal distribution of QoS operations is to have as many QoS actions
performed on the Catalyst switches as possible, saving the WAN/branch router valuable CPU
cycles. This is an especially critical consideration when deploying DoS/worm-mitigation
designs.
For example, some enterprises have deployed advanced QoS policies on their branch
switches and routers, only to have DoS/worm attacks originate from within the branch.
Remember, queuing will not engage on a switch unless its links are congestedand even
if it does, if the branch switch hands off 100 Mbps of (correctly queued) trafc to a branch
router, it more than likely will bring it down.
Thus, the following design principles for the campus-to-WAN handoff can help mitigate
these types of scenarios.
First, resist the urge to use a Gigabit Ethernet connection to the WAN aggregation router,
even if the router supports GE.
It is extremely unlikely that the WAN aggregator will be serving anywhere close to a
(combined) WAN circuit rate of 1 Gbps. Therefore, use one (or more) Fast Ethernet
connection on the distribution-layer Catalyst switch to connect to the WAG so that not only
is the aggregate trafc sent to the WAG limited (in 100 Mbps increments), but (because
congestion points now are pulled back into the Catalyst switch, thus forcing queuing to
engage on the FE switch port) the trafc also will be queued correctly within these limits
(of 100 Mbps increments).
For example, a WAN aggregation router is supporting two DS3 WAN connections (totaling
90 Mbps of WAN circuit capacity). In this case, the distribution-layer switch port connecting
to the WAG should be Fast Ethernet. Then, if more than 100 Mbps of trafc attempts to
traverse the WAN, the Catalyst switch will engage queuing on the switch port and aggressively
drop ows according to the dened application hierarchies. Only 100 Mbps of correctly
queued trafc will ever be handed off to the WAG.
In the case of a WAN aggregation router supporting over 100 Mbps of WAN circuits, as
in the case of a WAG running one or more OC-3 ports (at 155 Mbps each), multiple Fast
Ethernet connections can be used to connect to the WAG from the distribution-layer switch
to achieve the same net effect.
The point is to bring back, as much as possible, the choke point into Catalyst hardware and
engage hardware queuing there instead of overwhelming the software-based policing and
queuing policies within the WAN aggregation router.
Second, if the combined WAN circuit rate is signicantly below 100 Mbps, enable egress
shaping on the Catalyst switches (when supported).
If there is no hope of engaging queuing on the Catalyst switch because the combined WAN
circuit rates are far below those of Fast Ethernet (the minimum port speed of Catalyst
switches), enable shaping on platforms that support this feature. Such platforms include the
Catalyst 2970, 3560, 3750, and 4500.
422
In this manner, the Catalyst switch can hold back trafc and selectively drop (according to
dened policies) from ows that otherwise would ood the WAN/branch router.
For example, if a branch router is using two ATM-IMA T1 links (3 Mbps combined
throughput) to connect the branch to the WAN, the branch switch could be congured to
shape all WAN-destined trafc to 3 Mbps or could be congured to shape on a per-application
basis to smaller increments.
Refer to the queuing/dropping sections of these platforms in this chapter and Cisco IOS
documentation for additional guidance on enabling shaping.
Finally, if the combined WAN circuit rate is signicantly below 100 Mbps and the Catalyst
switch does not support shaping, enable egress policing (when supported).
If the Catalyst switch does not support shaping, egress policing is the next-best alternative
for this scenario.
For example, the Catalyst 3550 does not support shaping, but it does support up to eight
policers on all egress ports. Thus, it could still protect its branch router from being overwhelmed by policing on egress. Egress policing can be done on an aggregate level or on a
per-application basis.
Again, the objective is to discard, as intelligently as possible, trafc that will be dropped
inevitably anyway (by the WAN/branch router), but, whenever possible, to perform the
dropping within Catalyst hardware (instead of Cisco IOS Software).
Egress policers are congured in the same manner as ingress policers, but the direction
specied in the service-policy interface-conguration statement will be out, not in.
NOTE
The only Catalyst switch discussed in this chapter that did not support either shaping or
egress policing is the Catalyst 2950. Unfortunately, there is no way that the Catalyst 2950
can ofoad QoS from the branch router. If such functionality is required, a hardware
upgrade is advisable.

ABC, Inc., is a young, innovative company seeking to gain competitive advantage in its
industry vertical by strategic use of information technologies. As such, expectations from
its network infrastructure are considerable. These expectations include strict-priority
servicing of VoIP applications, and provisioning for high-quality videoconferencing and for
discrete/granular service levels for multiple classes of data applications. Furthermore,
ABC, Inc., expects to minimize network downtime from DoS and worm attacks by building
self-defending networks that leverage intelligent network services, such as quality of
service, intrusion detection, encryption, authentication, and other security technologies.
423
ABC, Inc., follows the Cisco recommended campus hierarchy of access, distribution, and
core layers and has deployed a mix of Catalyst switching platforms at these layers. These
switches include Catalyst 3550s and 3750s (at the access layer), 6500-Sup2s (within the
data center), 4500-Sup4s (within the distribution layer), and 6500-Sup720s (within the
distribution and core layers). Also, ABC, Inc., is connecting to Cisco 7200-NPE-G1 WAN
aggregation routers that are providing WAN services at OC-3 speeds. Figure 12-32 shows
the campus network topology.
Figure 12-32 Case Study: Campus Network with QoS Policies to Protect Voice, Video, and Data While Mitigating
DoS/Worm Attacks
Catalyst 6500-PFC3- Catalyst 6500-PFC3CORE-RIGHT
CORE-LEFT
10GE
Catalyst 4500-Sup4DISTRIBUTION-LEFT
Catalyst 6500-PFC3DISTRIBUTION-RIGHT
FE
Catalyst 6500-PFC2
-DATACENTER
FE
WAN Aggregator
ATM OC-3
Server Farms
Catalyst
Catalyst
3550
3750
IP Phones + PCs
IP Phones + PCs
Trust-DSCP + Queuing
Conditional Trust + Policing + Queuing
All links are Gigabit Ethernet unless otherwise noted.
Untrusted + Policing + Queuing

Per-User Microflow Policing
At this early phase in the network deployment, ABC, Inc., has basic networking congured
and wants to overlay QoS functionality. Following this phase, it will enable additional
security technologies.
ABC, Inc., has chosen to implement an Untrusted Server model within the data center
because it wants to administer QoS markings and policies on the network infrastructure
424
instead of on the application servers (which, at times, have been infected with viruses). The
Mission-Critical Data application is SAP, served off TCP ports 3200 to 3203 and 3600; the
Transactional Data application is Lotus Notes, served off TCP port 1352; and the Bulk Data
application is IMAP for E-mail, served off TCP ports 143 and 220. The company has
dedicated servers for these applications, but these servers sometimes are moved. Instead of
requiring the networking team to be notied for every server move, a blanket set of policies
will be deployed for these main applications on every data center switch port to correctly
mark and police these ows.
ABC, Inc., has deployed Cisco IP phones and is encouraging employees to make effective
use of videoconferencing applications to keep travel budgets low while maintaining productivity. Additionally, because the company has been hit hard with nearly a dozen worm
attacks within the past year, it has chosen to implement a Scavenger-class markdown/queuing strategy throughout the network to mitigate the effects of such ooding attacks. Finally,
because ABC, Inc., prefers data applications to receive QoS in both server-to-client and
client-to-server directions, it has chosen to deploy the Conditionally Trusted Endpoint: IP
Phone + PC Advanced model on all end-user access-layer switches. On the Catalyst 3550,
this model is being deployed through per-port/per-VLAN policers; on the Catalyst 3750, it
is being deployed by access lists that include voice VLAN subnet information.
All distribution-layer and core-layer links have been congured to trust DSCP markings.
Additionally, all distribution-layer Catalyst 6500-Sup720s (with PFC3s) are congured to
perform per-user microow policing as a second line of defense against spurious ows
from (non-data center access-layer switches). Queuing structures are dependant on
platform and line card capabilities, as detailed in the congurations.
Finally, ABC, Inc., has a series of Cisco 7200-NPE-G1 WAN aggregation routers to provide
WAN services. These WAN routers have ATM OC-3 (155-Mbps) links to their carrier and
home multiple ATM PVCs to their remote sites. Although these NPE-G1 routers have
Gigabit Ethernet interfaces, ABC, Inc., has chosen to connect to them using two Fast
Ethernet interfaces from the campus distribution-layer switches. In this manner, the amount
of trafc sent to WAN aggregators is limited to 200 Mbps, while intelligent queuing is
performedaccording to the enterprise-wide application hierarchywithin these physical
200-Mbps limits.
The QoS designs for the ABC campus network span six switches, but because one core
switch is a mirror image of the other, only ve congurations are detailed in the following
example. These include the QoS congurations for the Catalyst 6500 PFC2 data center
switch (see Example 12-76), the Catalyst 3550 access-layer switch (see Example 12-77),
the Catalyst 3750 access-layer switch (see Example 12-78), the left Catalyst 4500-Sup4
distribution-layer switch (see Example 12-79), the right Catalyst 6500-PFC3 distributionlayer switch (see Example 12-80), and the left Catalyst 6500-PFC3 core-layer switch (see
Example 12-81), of which the right Catalyst 6500-PFC3 core-layer switch is a mirror image.
Example 12-76 shows the QoS conguration for the Catalyst 6500 PFC2 data center
switch.
425
Example 12-76 Catalyst 6500-PFC2 (CatOS): Data Center Access Switch for Untrusted Servers (Advanced
Dual-Rate Policing Model): Case Study Example

!
#system
set system name C6500-PFC2-CATOS-DATACENTER
!
#!
#vtp
set vtp domain ABC-INC
set vtp mode transparent vlan
set vlan 1,200
!
#qos
set qos enable
! 1P2Q2T Global Definitions (for WS-X6K-S2U-MSFC2 GE Uplinks)
set qos map 1p2q2t tx 1 2 cos 0
! Assigns Net-Mgmt/Trans-Data to Q2T1
! Assigns Call-Sig/MC-Data to Q2T1
! Assigns Routing to Q2T2
! All other CoS-Queue mapping remain default
set qos wrr 1p2q2t 30 70
! Sets WRR weights to 30:70 for Q1:Q2
set qos txq-ratio 1p2q2t 30 40 30
! Sets sizes 30%,40%,30% for Q1,Q2,Q3/PQ
set qos wred 1p2q2t tx queue 1 40:80 80:100
! Sets WRED Thresholds for Q1
set qos wred 1p2q2t tx queue 2 70:80 80:100
! Sets WRED Thresholds for Q2
! 1P3Q1T Global Definitions (for WS-X6548-RJ-45 FE switch ports)
! Assigns Net-Mgmt/Trans-Data to Q3T1
! Assigns Call-Sig/MC-Data to Q3T1
! Assigns Video to Q3T1
set qos map 1p3q1t tx 3 cos 6
! Assigns IP Routing to Q3 (tail)
set qos map 1p3q1t tx 3 cos 7
! Assigns STP to Q3 (tail)
! All other CoS-Queue mapping remain default
set qos wrr 1p3q1t 5 25 70
! Sets WRR weights to 5:25:70 for Q1:Q2:Q3
set qos wred 1p3q1t tx queue 1 80:100 ! Sets WRED Thresholds for Q1 (80% to 100%)
set qos wred 1p3q1t tx queue 2 80:100 ! Sets WRED Thresholds for Q2 (80% to 100%)
set qos wred 1p3q1t tx queue 3 1:80
70:80 ! Sets WRED Thresholds for Q3 (70% to 80%)
set qos policed-dscp-map 1:1
set qos policed-dscp-map 0,8,25:8
! Normal markdown of 0 and 25 set to CS1
set qos policed-dscp-map 10,12:12
! Normal markdown of AF11 set to AF12
continues
426
Dual-Rate Policing Model): Case Study Example (Continued)

set qos policed-dscp-map 18,20:20
! Normal markdown of AF21 set to AF22
<remaining policed-dscp-map (normal rate) DSCP values are mapped to themselves>
set qos policed-dscp-map excess-rate 0:0

set qos policed-dscp-map excess-rate 8,10,18:8
! Excess markdown of AF11 and AF12 set to CS1
<remaining policed-dscp-map (excess rate) DSCP values are mapped to themselves>
set qos policer aggregate SAP rate 15000 policed-dscp erate 32000000
policed-dscp burst 8000
eburst
32000
8 eburst
32000
! SAP traffic is (single-rate) policed to 15 Mbps
set qos policer aggregate LOTUS rate 25000 policed-dscp erate 35000
eburst
8000
8 eburst
8000
! Lotus traffic is (dual-rate) policed to 25 Mbps and 35 Mbps
set qos policer aggregate IMAP rate 40000 policed-dscp erate 50000
eburst
8000
8 eburst
8000
! IMAP traffic is (dual-rate) policed to 40 Mbps and 50 Mbps
set qos policer aggregate DATA rate 1000 policed-dscp erate 32000000
eburst
32000
8 eburst
32000
! All other data is (single-rate) policed to 1 Mbps
clear qos acl all
#UNTRUSTED-SERVER
set qos acl ip UNTRUSTED-SERVER dscp 25 aggregate SAP tcp any range 3200 3203 any
set qos acl ip UNTRUSTED-SERVER dscp 25 aggregate SAP tcp any eq 3600 any
! Identifies SAP traffic by TCP source ports 3200-3203 and 3600
set qos acl ip UNTRUSTED-SERVER dscp 18 aggregate LOTUS tcp any eq 1352 any
! Identifies Lotus traffic by TCP source port 1352
set qos acl ip UNTRUSTED-SERVER dscp 10 aggregate IMAP tcp any any eq 143
set qos acl ip UNTRUSTED-SERVER dscp 10 aggregate IMAP tcp any any eq 220
! Identifies IMAP traffic by TCP source port 143 and 220
set qos acl ip UNTRUSTED-SERVER dscp 0 aggregate DATA ip any any
! ACL to catch all other IP traffic
#
commit qos acl all
! Commits all ACLs to PFC
!
# default port status is enable
427
Dual-Rate Policing Model): Case Study Example (Continued)

!
!
#module 1 : 2-port 1000BaseX Supervisor
clear trunk 1/1 1-199,201-1005,1025-4094
set trunk 1/1 on dot1q 200
clear trunk 1/2 1-199,201-1005,1025-4094
set trunk 1/2 on dot1q 200
set
port
qos
1/1-2
trust-dscp
commit
qos
set
porttrust
qos 1/1-2
trust
!
#module 3 : 48-port 10/100BaseTX Ethernet
set vlan 200 3/1-48
set qos acl map UNTRUSTED-SERVER 3/1-48
!
! Trunks Data Center VLAN

! Uplink ports Trust DSCP
! Data Center VLAN

! ACL attached to all access-ports
Example 12-77 shows the QoS conguration for the (left) Catalyst 3550 access-layer
switch.
Example 12-77 Catalyst 3550 Access Switch for IP Phones + PCs (Advanced Model): Case Study
Example
!
hostname CAT3550-ACCESS-LEFT
!
vtp domain ABC-INC
vtp mode transparent
! Excess DVLAN traffic marked 0, AF11, AF21, CS3, DSCP 25 and AF41
! will be remarked to Scavenger (CS1)
mls qos
! Enables QoS Globally
!
match ip dscp 46
! DSCP EF (voice)
class-map match-all VVLAN-VOICE
match vlan 110
! VLAN 110 is VVLAN
match class-map VOICE
! Matches VVLAN DSCP EF
!
class-map match-any CALL-SIGNALING
! Need 'match-any' here
match ip dscp 26
match ip dscp 24
class-map match-all VVLAN-CALL-SIGNALING
match vlan 110
! VLAN 110 is VVLAN
match class-map CALL-SIGNALING
! Matches VVLAN AF31/CS3
!
class-map match-all ANY
match access-group name ANY
! Workaround ACL
continues
428
Example (Continued)
class-map match-all VVLAN-ANY
match vlan 110
! VLAN 110 is VVLAN
10
match class-map ANY
!
class-map match-all PC-VIDEO
match access-group name PC-VIDEO
class-map match-all DVLAN-PC-VIDEO
match vlan 10
! VLAN 10 is DVLAN
match class-map PC-VIDEO
! Matches PC IP/VC or SoftPhone
!
class-map match-all MISSION-CRITICAL-DATA
match access-group name MISSION-CRITICAL-DATA
class-map match-all DVLAN-MISSION-CRITICAL-DATA
match vlan 10
! VLAN 10 is DVLAN
match class-map MISSION-CRITICAL-DATA
! Matches DVLAN MC-Data
!
class-map match-all TRANSACTIONAL-DATA
match access-group name TRANSACTIONAL-DATA
class-map match-all DVLAN-TRANSACTIONAL-DATA
match vlan 10
! VLAN 10 is DVLAN
match class-map TRANSACTIONAL-DATA
! Matches DVLAN Transactional
!
class-map match-all BULK-DATA
match access-group name BULK-DATA
class-map match-all DVLAN-BULK-DATA
match vlan 10
! VLAN 10 is DVLAN
match class-map BULK-DATA
! Matches DVLAN Bulk Data
!
class-map match-all DVLAN-ANY
match vlan 10
! VLAN 10 is DVLAN
match class-map ANY
! Matches all other DVLAN traffic
!
!
policy-map IPPHONE+PC-ADVANCED
class VVLAN-VOICE
! DSCP EF (Voice)
set ip dscp 46
class VVLAN-CALL-SIGNALING
set ip dscp 24
class VVLAN-ANY
set ip dscp 0
class DVLAN-PC-VIDEO
set ip dscp 34
class DVLAN-MISSION-CRITICAL-DATA
429
Example (Continued)
set ip dscp 25
class DVLAN-TRANSACTIONAL-DATA
set ip dscp 18
class DVLAN-BULK-DATA
set ip dscp 10
class DVLAN-ANY
set ip dscp 0
!
!
description ACCESS-EDGE IP PHONE + PC ADVANCED MODEL
switchport access vlan 10
! DVLAN
switchport mode dynamic desirable
switchport voice vlan 110
! VVLAN
no ip address
! Conditional Trust
service-policy input IPPHONE+PC-ADVANCED
! Attaches policy
wrr-queue bandwidth 5 25 70 1
wrr-queue cos-map 1 1
! Scavenger/Bulk is assigned to Q1
! Best Effort is assigned to Q2
wrr-queue cos-map 3 2 3 4 6 7
! CoS 2,3,4,6,7 are assigned to Q3
! Voice is assigned to Q4 (PQ)
priority-queue out
! Enables Q4 as PQ
spanning-tree portfast
!
<repeated for all FE switchports>
!
!
interface GigabitEthernet0/1
description L2 UPLINK TO DISTRIBUTION CAT4500-SUP4-LEFT
switchport trunk encapsulation dot1q
! Trunks DVLAN + VVLAN
switchport trunk allowed vlan 10,110
switchport mode trunk
no ip address
! Trusts DSCP
mls qos trust dscp
wrr-queue bandwidth 5 25 70 1
wrr-queue queue-limit 5 25 40 30
! Tunes buffers to 5% for Q1, 25% for Q2, 40% for Q3 and 30% for Q4
continues
430
Example (Continued)
wrr-queue random-detect max-threshold 1 40 100
! Scavenger/Bulk is assigned to Q1
! Best Effort is assigned to Q2
! CoS 2,3,4,6,7 are assigned to Q3
wrr-queue cos-map 3 2 3 4 6 7
! Voice is assigned to Q4 (PQ)
wrr-queue dscp-map 2 10 48 56
! Maps Bulk Data (10), Routing (48) and Spanning Tree (Internal DSCP 56)
! to WRED threshold 2 of their respective queues all other DSCP values
! are mapped (by default) to WRED threshold 1 of their respective queues
! Enables Q4 as PQ
priority-queue out
!
<repeated for GigabitEthernet0/2 Uplink to Distribution Cat4500-Sup4-Left>
!
!
ip access-list standard ANY
permit any
!
ip access-list extended PC-VIDEO
! IP/VC or SoftPhone
permit udp any any range 16384 32767
!
ip access-list extended MISSION-CRITICAL-DATA
permit tcp any any range 3200 3202
! SAP
permit tcp any any eq 3600
! SAP
! Softphone SCCP
!
ip access-list extended TRANSACTIONAL-DATA
! Lotus
!
ip access-list extended BULK-DATA
! IMAP
! IMAP
!
Example 12-78 shows the QoS conguration for the (right) Catalyst 3750 access-layer
switch.
Example
!
hostname CAT3750-ACCESS-RIGHT
!
431
Example (Continued)
vtp domain ABC-INC
! Excess DVLAN traffic marked 0, AF11, AF21, CS3, DSCP 25 and AF41
! will be remarked to Scavenger (CS1)
!
mls qos srr-queue output cos-map queue 1
3 threshold 3 5
0
! Maps CoS 5 to Queue 1 Threshold 3 (Voice gets all of Queue 1)
! Maps CoS 3 to Queue 2 Threshold 2
! Maps CoS 0 to Queue 3 Threshold 3 (Best Efforts gets all of Q3)
! Maps CoS 1 to Queue 4 Threshold 3 (Scavenger/Bulk gets all of Q4)
!
! Maps DSCP EF (Voice) to Queue 1 Threshold 3
mls qos srr-queue output dscp-map queue 2 threshold 1 16 18 20 22 25 32 34 36
! Maps DSCP CS2 (Net-Mgmt/Transactional) to Queue 2 Threshold 1
! Maps DSCP AF21, AF22, AF23 (Transactional Data) to Queue 2 Threshold 1
! Maps DSCP 25 (Mission-Critical Data) to Queue 2 Threshold 1
! Maps DSCP CS4 (Streaming Video) to Queue 2 Threshold 1
! Maps DSCP AF41, AF42 (Interactive-Video) to Queue 2 Threshold 1
! Maps DSCP AF43 (Interactive-Video) to Queue 2 Threshold 1
! Maps DSCP CS3 and DSCP AF31 (Call-Signaling) to Queue 2 Threshold 2
! Maps DSCP CS6 and CS7 (Network/Internetwork) to Queue 2 Threshold 3
! Maps DSCP 0 (Best Effort) to Queue 3 Threshold 3
! Maps DSCP CS1 (Scavenger) to Queue 4 Threshold 1
! Maps DSCP AF11, AF12, AF13 (Bulk Data) to Queue 4 Threshold 3
!
mls qos
!
continues
432
Example (Continued)
class-map match-all VVLAN-VOICE
match access-group name VVLAN-VOICE
class-map match-all VVLAN-CALL-SIGNALING
match access-group name VVLAN-CALL-SIGNALING
class-map match-all VVLAN-ANY
match access-group name VVLAN-ANY
!
class-map match-all DVLAN-PC-VIDEO
match access-group name DVLAN-PC-VIDEO
class-map match-all DVLAN-MISSION-CRITICAL-DATA
match access-group name DVLAN-MISSION-CRITICAL-DATA
class-map match-all DVLAN-TRANSACTIONAL-DATA
match access-group name DVLAN-TRANSACTIONAL-DATA
class-map match-all DVLAN-BULK-DATA
match access-group name DVLAN-BULK-DATA
!
policy-map IPPHONE+PC-ADVANCED
class VVLAN-VOICE
set ip dscp 46
! DSCP EF (Voice)
class VVLAN-CALL-SIGNALING
set ip dscp 24
class VVLAN-ANY
set ip dscp 0
class DVLAN-PC-VIDEO
set ip dscp 34
class DVLAN-MISSION-CRITICAL-DATA
set ip dscp 25
class DVLAN-TRANSACTIONAL-DATA
set ip dscp 18
class DVLAN-BULK-DATA
set ip dscp 10
class class-default
set ip dscp 0
!
433
Example (Continued)
!
interface FastEthernet1/0/1
description ACCESS-EDGE IP PHONE + PC ADVANCED MODEL
! DVLAN
switchport access vlan 20
! VVLAN
switchport voice vlan 120
no ip address
srr-queue bandwidth share 1 70 25 5
! Q1 is PQ; Q2 gets 70% of remaining BW; Q3 gets 25% and Q4 gets 5%
! Q1 is limited to 30%
srr-queue bandwidth shape 30 0 0 0
priority-queue out
! Conditional Trust
! Attaches policy to interface
service-policy input IPPHONE+PC-ADVANCED
no mdix auto
spanning-tree portfast
!
<repeated for all FE switchports>
!
interface GigabitEthernet1/0/1
description L2 UPLINK TO DISTRIBUTION CAT4500-SUP4-LEFT
! Trunks DVLAN and VVLAN
no ip address
srr-queue bandwidth share 1 70 25 5
! Q1 is PQ; Q2 gets 70% of remaining BW; Q3 gets 25% and Q4 gets 5%
! Q1 is shaped to 30%
srr-queue bandwidth shape 30 0 0 0
priority-queue out
! Trusts DSCP
mls qos trust dscp
!
<repeated for GigabitEthernet1/0/2 Uplink to Distribution Cat4500-Sup4>
!
!
ip access-list extended VVLAN-VOICE
permit udp 10.1.110.0 0.0.0.255 any range 16384 32767 dscp ef
!
ip access-list extended VVLAN-CALL-SIGNALING
permit tcp 10.1.120.0 0.0.0.255 any range 2000 2002 dscp af31
permit tcp 10.1.120.0 0.0.0.255 any range 2000 2002 dscp cs3
! Call-Signaling is matched by VVLAN subnet and DSCP AF31 or CS3
!
ip access-list extended VVLAN-ANY
permit ip 10.1.120.0 0.0.0.255 any
!
ip access-list extended DVLAN-PC-VIDEO
! DVLAN IP/VC
permit udp any any range 16384 32767
!
continues
434
Example (Continued)
ip access-list extended DVLAN-MISSION-CRITICAL-DATA
!
ip access-list extended DVLAN-TRANSACTIONAL-DATA
!
ip access-list extended DVLAN-BULK-DATA
!
! SAP
! SAP
! SoftPhone SCCP
! IMAP
! IMAP
! IMAP
Example 12-79 shows the QoS conguration for the (left) Catalyst 4500-Sup4 distributionlayer switch.
Example 12-79 Catalyst 4500-Sup4 Distribution Switch Case Study Example
!
hostname C4500-SUP4-DIST-LEFT
!
! Optional: Enables DBL to mark IP ECN bits
qos dbl
! Globally enables DBL
! Maps Best Effort to Queue 2
qos map dscp 16 18 20 22 24 25 26 32 to tx-queue 4
! Maps DSCP CS2 (Net-Mgmt) and AF21/AF22/AF23 (Transactional) to Q4
! Maps DSCP CS3 and AF31 (Call-Signaling) and DSCP 25 (MC Data) to Q4
! Maps DSCP CS4 (Str-Video) to Q4
qos map dscp 34 36 38 to tx-queue 4
! Maps DSCP AF41/AF42/AF43 (Int-Video) to Q4
! All other DSCP-to-Queue Mappings are remain at default
qos
!
vtp domain ABC-INC
!
!
policy-map DBL
class class-default
! Enables DBL on all traffic flows
dbl
!
!
description L3 UPLINK TO CORE 6500-PFC3-LEFT
no switchport
ip address 10.1.230.2 255.255.255.252
service-policy output DBL
qos trust dscp
! Trusts DSCP
Example 12-79 Catalyst 4500-Sup4 Distribution Switch Case Study Example (Continued)
tx-queue 1
bandwidth percent
tx-queue 2
bandwidth percent
tx-queue 3
bandwidth percent
priority high
shape percent 30
tx-queue 4
bandwidth percent
! Q1 gets 5%
25
! Q2 gets 25%
30
! Enables Q3 as PQ
! PQ gets 30%
! Shapes PQ to 30%
40
! Q4 gets 40%
!
<repeated for GigabitEthernet1/2 Uplink to Core 6500-PFC3-Right>
!
!
description L3 FASTETHERNET LINK TO WAG 7200-NPEG1
no switchport
ip address 10.1.220.1 255.255.255.252
! Trusts DSCP
qos trust dscp
tx-queue 3
priority high
! PQ gets 30%
shape percent 30
! Shapes PQ to 30%
!
<repeated for FastEthernet2/2 Link to WAN Aggregator>
!
!
description L2 DOWNLINK TO DATA CENTER 6500-PFC2
switchport trunk allowed vlan 200
! Trusts DSCP
qos trust dscp
tx-queue 1
! Q1 gets 5%
bandwidth percent 5
tx-queue 2
! Q2 gets 25%
tx-queue 3
! Enables Q3 as PQ
priority high
! PQ gets 30%
shape percent 30
! Shapes PQ to 30%
tx-queue 4
! Q4 gets 40%
!
<repeated for GigabitEthernet3/2 Downlink to Access-Layer Cat3550>
!
435
436
Example 12-80 shows the QoS conguration for the (right) Catalyst 6500-PFC3
distribution-layer switch.
Example 12-80 Catalyst 6500-PFC3 IOS Distribution Switch (with Per-User Microow Policing): Case Study
Example
!
hostname CAT6500-PFC3-IOS-DIST-RIGHT
!
!
mls qos map policed-dscp normal-burst 0 24 26 34 36 to 8
! Excess traffic marked 0,CS3,AF31,AF41 or AF42 will be remarked to CS1
mls qos ! Enables QoS Globally
!
!
class-map match-all VOIP
match ip dscp ef
class-map match-all INTERACTIVE-VIDEO
match ip dscp af41 af42
class-map match-all CALL-SIGNALING
match ip dscp cs3 af31
class-map match-all BEST-EFFORT
match ip dscp default
!
!
policy-map PER-USER-POLICING
class VOIP
police flow mask src-only 128000 8000
conform-action transmit exceed-action drop
! No source can send more than 128k worth of Voice traffic
! Excess IP/VC traffic from any source is marked down to CS1
! Excess Call-Signaling from any source is marked down to CS1
class BEST-EFFORT
! Excess PC Data traffic from any source is marked down to CS1
!
!
! WS-X6408A-GBIC (1P2Q2T)
description L3 UPLINK TO CORE 6500-PFC3-LEFT
ip address 10.1.230.6 255.255.255.252
wrr-queue bandwidth 30 70
wrr-queue queue-limit 40 30
437
Example (Continued)
wrr-queue random-detect min-threshold 1 40 40
! Maps Scavenger/Bulk to Q1 WRED T1
wrr-queue cos-map 1 1 1
! Maps Best Effort to Q1 WRED T2
! Maps CoS 2,3,4 to Q2 WRED T1
wrr-queue cos-map 2 1 2 3 4
! Maps CoS 6 and 7 to Q2 WRED T2
wrr-queue cos-map 2 2 6 7
! Trusts DSCP
mls qos trust dscp
!
<repeated for GigabitEthernet1/2 Uplink to Core 6500-PFC3-Right>
!
!
! WS-X6548-RJ-45 (1P3Q1T)
description L3 FASTETHERNET LINK TO WAG 7200-NPEG1
ip address 10.1.220.10 255.255.255.252
wrr-queue bandwidth 5 25 70
wrr-queue random-detect min-threshold 1 80
! Sets the Min WRED Threshold for Q1T1 to 80%
! All other WRED Thresholds are left at default (100%)
wrr-queue cos-map
3 1 3
2 1
3 2
4 3
6 4
7
! Maps CoS 2,3,4,6 and 7 to Q3 WRED T1
wrr-wrr-queue
cos-map
! Trusts DSCP
mls qos trust dscp
!
<repeated for FastEthernet2/2 Link to WAN Aggregator>
!
!
description L2 DOWNLINK TO ACCESS-LAYER C3550 - LEFT
no ip address
continues
438
Example (Continued)
,
! Trusts DSCP
mls qos trust dscp
switchport
! Trunks DVLAN and VVLAN
service-policy input PER-USER-POLICING
! Attaches Per-User Policer
!
<repeated for GigabitEthernet3/1 Downlink to Data Center C6500-PFC2>
<except no "service-policy input PER-USER-POLICING" for Data Center Downlink>
!
Example 12-81 shows the QoS conguration for the left Catalyst 6500-PFC3 core-layer
switch, of which the right Catalyst 6500-PFC3 core-layer switch is a mirror image.
Example 12-81 Catalyst 6500-PFC3 IOS Core Switch: Case Study Example
!
hostname CAT6500-PFC3-IOS-CORE-LEFT
!
mls qos
!
!
interface Port-channel1
description 40GE CORE BACKBONE
ip address 10.1.240.1 255.255.255.252
mls qos trust dscp
!
!
interface TenGigabitEthernet1/1
! WS-X6704-10GE (1P7Q8T)
no ip address
wrr-queue bandwidth 5 25 20 20 20 5 5
wrr-queue queue-limit 5 25 10 10 10 5 5
! Allocates 10% to Q5, 5% to Q6 and 5% to Q7
439
Example 12-81 Catalyst 6500-PFC3 IOS Core Switch: Case Study Example (Continued)
wrr-queue random-detect min-threshold 1 80 100 100 100 100 100 100 100
wrr-queue random-detect max-threshold 1 100 100 100 100 100 100 100 100
wrr-queue random-detect max-threshold 2 100 100 100 100 100 100 100 100
! All other WRED Thresholds remain at their default values (100%)
wrr-queue random-detect 4
! Maps Scavenger/Bulk to Q1
! Maps Best Effort to Q2
! Maps Video to Q3
! Maps Net-Mgmt/Transactional to Q4
! Maps Call-Signaling/MC-Data to Q5
! Maps IP Routing to Q6
! Maps STP to Q7
! Trusts DSCP
mls qos trust dscp
channel-group 1 mode on
!
<repeated three more times on interfaces TenGigabitEthernet1/2-4>
!
!
description L3 DOWNLINK TO DISTRIBUTION 4500-SUP4-LEFT
ip address 10.1.230.1 255.255.255.252
continues
440
Example 12-81 Catalyst 6500-PFC3 IOS Core Switch: Case Study Example (Continued)
! Trusts DSCP
mls qos trust dscp
!
<repeated for GigabitEthernet2/2 Downlink to Distribution 6500-PFC3-Right>
!
Summary
This chapter began with establishing the case for campus QoS by way of three main QoS
design principles: The rst is that applications should be classied and marked as close to
their sources as technically and administratively feasible. The second is that unwanted
trafc ows should be policed as close to their sources as possible. The third is that QoS
always should be performed in hardware, instead of software, whenever a choice exists.
Furthermore, it was emphasized that the only way to provide service guarantees is to enable
queuing at any node that has the potential for congestion, including campus uplinks and
downlinks.
A proactive approach to mitigating DoS/worm ooding attacks within campus environments
was overviewed. This approach focuses on access-edge policers that meter trafc rates
received from endpoint devices; when these exceed specied watermarks (at which point
they no longer are considered normal ows), these policers mark down excess trafc to
Scavenger (DSCP CS1). These policers are coupled with queuing policies throughout the
enterprise that provision for a less-than best-effort Scavenger class on all links. In this manner, legitimate trafc bursts are not affected, but DoS/worm-generated trafc signicantly
is mitigated.
Common endpoints were overviewed and classied into three main groups: trusted endpoints,
untrusted endpoints, and conditionally trusted endpoints. Untrusted endpoints were subdivided into two smaller models: untrusted PCs and untrusted servers. Similarly, conditionally
trusted endpoints were subdivided into two models: basic and advanced.
Following these access-edge model denitions, platform-specic recommendations were
given on to how to implement these access-edge models on Cisco Catalyst 2950, 2970,
3550, 3560, 3750, 4500, and 6500 series switches. Platform-specic limitations, caveats,
and nerd knobs were highlighted to tailor each model to each platforms unique feature sets.
All congurations were presented in cong mode to continually highlight what platform
was being discussed. Furthermore, many relevant verication commands were discussed in
detail (in context) to illustrate how and when these can be used effectively when deploying
QoS within the campus.
Further Reading
441
Recommendations also were given on how to congure queuing on a per-platform, perline-card basis. These recommendations included conguring 1P3Q1T queuing on the
Catalyst 2950, conguring 1P3Q2T queuing on the Catalyst 3550, conguring 1P3Q3T
queuing on the Catalyst 2970/3560/3750, and conguring 1P3Q1T queuing (+ DBL) on the
Catalyst 4500. For the Catalyst 6500, line cardspecic queuing structures were examined
in detail, including CatOS and Cisco IOS congurations for conguring 2Q2T, 1P2Q1T,
1P2Q2T, 1P3Q1T, 1P3Q8T, and 1P7Q8T queuing.
Following this, the Catalyst 6500 PFC3s per-user microow policing feature was
discussed in the context of how it can be leveraged to provide a second line of policing
defense at the distribution layer.
Finally, campus-to-WAN/VPN handoff considerations were examined. It was recommended
that you rst resist the urge to use a Gigabit Ethernet connection to the WAN aggregation
router, even if the router supports GE. Second, if the combined WAN circuit rate is significantly below 100 Mbps, enable egress shaping on the Catalyst switches (when supported).
Third, if the combined WAN circuit rate is signicantly below 100 Mbps and the Catalyst
switch does not support shaping, enable egress policing (when supported).
The design chapter concluded with a case study of a ctitious enterprise, to illustrate how
these various platform-specic recommendations can be brought together in an end-to-end
campus QoS design for protecting voice, video, and critical data applications while
mitigating DoS/worm attacks.
Further Reading
Standards:
IPv6 Headers: http://www.ietf.org/rfc/rfc2474.
RFC 2597, Assured Forwarding PHB Group: http://www.ietf.org/rfc/rfc2597.
RFC 3246, An Expedited Forwarding PHB (Per-Hop Behavior):

http://www.ietf.org/rfc/rfc3246.

RFC 3168, The Addition of Explicit Congestion Notication (ECN) to IP:
http://www.ietf.org/rfc/rfc3168.
Books:
Flannagan, Michael, Richard Froom, and Kevin Turek. Cisco Catalyst QoS: Quality
of Service in Campus Networks. Indianapolis: Cisco Press, 2003.
442
Cisco Catalyst documentation:

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat2950/12119ea1/
2950scg/swqos.htm.

http://www.cisco.com/univercd/cc/td/doc/product/lan/c3550/12119ea1/
3550scg/swqos.htm.

2970scg/swqos.htm.

3750scg/swqos.htm.
Conguring QoS on the Catalyst 4500 (Cisco IOS Release 12.2[18]EW):

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat4000/12_2_18/
cong/qos.htm.
Conguring QoS on the Catalyst 6500 (Cisco CatOS Release 8.2):

confg_gd/qos.htm.
Conguring Automatic QoS on the Catalyst 6500 (Cisco CatOS Release 8.2):
confg_gd/autoqos.htm.
Conguring QoS on the Catalyst 6500 (Cisco IOS Release 12.2[17]SX):

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat6000/122sx/
swcg/qos.htm.
PA R T
IV
WAN QoS Design

Part IV of this book provides an in-depth discussion of private-WAN QoS design. QoS considerations and
detailed designs are presented for WAN aggregation routers connecting campus networks to various
WAN media, such as leased-lines, Frame Relay, ATM and ATM-to-Frame Relay Service-Interworking.
Additionally, QoS considerations and designs unique to branch routers are examined, which include using
Network-Based Application Recognition (NBAR) for known-worm policing. While primarily addressing
private-WAN QoS design, many of these recommendations are also applicable to VPN QoS designs
(discussed in Part V) where traditional private-WAN media is used for VPN access.
Chapter 13
Chapter 14
This chapter discusses WAN QoS considerations and designs, including the following:
Slow-speed ( 768 kbps) WAN link design

Medium-speed (768 kbps to T1/E1 speed) WAN link design
High-speed (> T1/E1 speed) WAN link design
Additionally, these designs are applied to specic Layer 2 WAN media, including the
following:
Leased lines
Frame Relay
ATM
ISDN
CHAPTER
13

A fundamental principle of economics states that the more scarce a resource is the more
efciently it should be managed. In an enterprise network infrastructure, bandwidth is
the prime resource and also is the scarcest (and, likewise, most expensive) over the WAN.
Therefore, the case for efcient bandwidth optimization using QoS technologies is strongest
over the WAN, especially for enterprises that are converging their voice, video, and data
networks.
The design principles described in this chapter apply primarily to Layer 2 WANs, such as
leased lines, Frame Relay, and ATM (including ATM-to-Frame Relay Service Interworking). However, many service providers use these Layer 2 WAN technologies to access
Layer 3 VPN services. Therefore, many of the design principles and examples presented
in this chapter also apply to such VPN access scenarios.
This chapter provides design guidance for enabling QoS over the WAN. It is important to
note that the recommendations in this chapter are not autonomous. They are critically
dependent on the recommendations discussed in Chapter 12, Campus QoS Design.
Where Is QoS Needed over the WAN?

Within typical WAN environments, routers play one of two roles: a WAN aggregator or
a branch router. In some very complex WAN models, enterprises might have distributed
WAN aggregators to cover regional branches, but the role of such middle-tier routers is
not signicantly different from that of a WAN aggregator located at a campus edge. This
chapter focuses on WAN edge recommendationsprimarily for WAN aggregator routers,
but these correspondingly apply to the WAN edge designs of branch routers. QoS policies
required on WAN edges are shown in Figure 13-1.
448
Chapter 13: WAN Aggregator QoS Design
Figure 13-1 Where Is QoS Needed over the WAN?

Campus
Distribution
Switches
WAN Aggregator
WAN
WAN Edges
LAN Edges
Queuing/Dropping/Shaping/
Link-Efficiency Policies for
Campus-to-Branch Traffic
Chapter 14, Branch Router QoS Design, discusses additional QoS considerations and
designs unique to branch routers.
WAN Edge QoS Design Considerations

QoS policies required on WAN aggregators include queuing, shaping, selective dropping,
and link-efciency policies in the outbound direction of the WAN link. Trafc is assumed
to be correctly classied and marked (at Layer 3) before WAN aggregator ingress. Remember, Layer 3 markings (preferably DSCP) are media independent and traverse the WAN
media, whereas Layer 2 CoS is lost when the media switches from Ethernet to WAN media.
Several factors must be kept in mind when designing and deploying QoS polices on WAN
edges. Some of these considerations were introduced in earlier chapters. They are reemphasized here to underscore their importance to the context of the WAN QoS designs
that follow.
Software QoS
Unlike LAN (Catalyst) queuing, which is done in hardware, WAN edge QoS is performed
within Cisco IOS Software. If the WAN aggregator is homing several hundred remote
branches, the collective CPU required to administer complex QoS policies might be more
than some older devices can provide.
449
The main point to keep in mind is that QoS entails a marginal CPU load. WAN topologies
and QoS policies should be designed to limit the average CPU utilization of the WAN
aggregator to 75 percent (or lower) because this leaves cycles available to respond efciently to routing updates.
Bandwidth Provisioning for Best-Effort Trafc

As discussed previously, the Best-Effort class is the default class for all data trafc. Only if
an application has been selected for preferential or deferential treatment is it removed from
the default class. Because many enterprises have several hundreds, if not thousands, of data
applications running over their networks, adequate bandwidth must be provisioned for this
class as a whole to handle the sheer volume of applications that default to it. It is recommended that at least 25 percent of a WAN links bandwidth be reserved for the default BestEffort class.
Bandwidth Provisioning for Real-Time Trafc

Not only does the Best-Effort class of trafc require special bandwidth-provisioning
consideration, but the Real-Time class does as well. The amount of bandwidth assigned
to the Real-Time class is variable; however, if too much trafc is assigned to Real-Time
(strict-priority/low-latency) queuing, the overall effect is a dampening of QoS functionality
for data applications.
The goal of convergence cannot be overemphasized: to enable voice, video, and data to
coexist transparently on a single network. When real-time applications (such as voice or
interactive-video) dominate a WAN link, data applications uctuate signicantly in their
response times, destroying the transparency of the converged network.
Cisco Technical Marketing testing has shown a signicant decrease in data application
response times when Real-Time trafc exceeds one-third of a links bandwidth capacity.
Cisco IOS Software allows the abstraction (and, thus, conguration) of multiple LLQs.
Extensive testing and production-network customer deployments have shown that limiting
the sum of all LLQs to 33 percent is a conservative and safe design ratio for merging realtime applications with data applications.
Furthermore, it should be kept in mind that if VoIP trafc is set to dominate a link via lowlatency queuing (which is essentially strict-priority FIFO queuing), VoIP actually could
negatively impact other VoIP trafc because of extensive FIFO queuing. This easily could
result in excessive serialization delays ( 10 ms per hop) on even medium-speed links
(T1/E1 links) where serialization delays ordinarily would not even be a consideration.
(Serialization delays are discussed in more detail in the next section.) Such excessive
serialization delays from VoIP LLQ overprovisioning would increase VoIP jitter and, thus,
decrease overall call quality.
450
NOTE
The 33-percent limit for the sum of all LLQs is simply a best-practice design recommendation; it is not a mandate. In some cases, specic business objectives cannot be met while
holding to this recommendation. In such cases, enterprises must provision according to
their detailed requirements and constraints. However, it is important to recognize the tradeoffs involved with overprovisioning LLQ trafc in respect to the negative performance
impact on data application response times.
Serialization
Serialization delay refers to the nite amount of time it takes to clock a frame onto the
physical media. Within the campus, this time is so innitesimal that it is completely
immaterial. Over the WAN, however, lower link speeds can cause sufcient serialization
delay to adversely affect real-time streams, such as Voice or Interactive-Video.
Serialization delays are variable because they depend not only on the line rate of the link
speed, but also on the size of the packet being serialized. Variable (network) delay also is
known as jitter. Because the end-to-end one-way jitter target has been set as 30 ms, the typical per-hop serialization delay target is 10 ms (which allows for up to three intermediate
hops per direction of VoIP trafc ow). This 10 ms per-hop target leads to the recommendation that a link fragmentation and interleaving (LFI) tool (either MLP LFI or FRF.12) be
enabled on links with speeds at or below 768 kbps (this is because the serialization delay
of a maximum-size Ethernet packet1500 bytestakes more than 10 ms to serialize at
768 kbps and below). Naturally, LFI tools need to be enabled on both ends of the link.
When deploying LFI tools, it is recommended that the LFI tools be enabled during a scheduled downtime. Assuming that the network administrator is within the enterprises campus,
it is recommended that LFI be enabled on the branch router rst (which is on the far end of
the WAN link) because this generally takes the WAN link down. Then the administrator can
enable LFI on the WAN aggregator (the near end of the WAN link), and the link will come
back up. Otherwise, if the administrator enables LFI on the WAN aggregator rst, the link
will go down, along with any in-band management access to the branch router. In such a
case, the administrator would need to remove LFI from the WAN aggregator (bringing the
link back up), enable LFI on the branch router, and then re-enable LFI on the WAN
aggregator.
Additionally, as pointed out in Chapter 5, Congestion-Management Tools, because trafc
assigned to the LLQ escapes fragmentation, it is recommended that Interactive-Video not
be deployed on slow-speed links; the large Interactive-Video packets (such as 1500-byte
full-motion I-Frames) could cause serialization delays for smaller Interactive-Video
packets. Interactive-Video trafc patterns and network requirements are overviewed in
Chapter 2, QoS Design Overview.
451
IP RTP Header Compression

Compressing IP, UDP, and RTP headers (cRTP) for VoIP calls can result in signicant
bandwidth gains over WAN links. However, it is important to realize that cRTP is one of
the most CPU-intensive features within the Cisco IOS Software QoS toolset. Therefore,
it is recommended that cRTP be used primarily on slow-speed ( 768 kbps) links with a
careful eye on CPU levels (especially for WAN aggregators that home a large number of
remote branches).
Tx-ring Tuning
Newer versions of Cisco IOS Software automatically size the nal interface output buffer
(Tx-ring) to optimal lengths for Real-Time applications, such as Voice or Video. On some
older versions of Cisco IOS Software, Tx-rings might need to be reduced on slow-speed
links to avoid excessive serialization delay.
To determine the value of the Tx-ring on an interface, use the variation of the show
controllers command shown in Example 13-1.
Example 13-1 Displaying the Tx-ring Value with the show controllers Command
WAG-7206-Left#show controllers Serial 1/0 | include tx_limited
tx_underrun_err=0, tx_soft_underrun_err=0, tx_limited=1(64)
64
WAG-7206-Left#
The value within the parentheses following the tx_limited keyword reects the value of the
Tx-ring. In this particular example, the Tx-ring is set to 64 packets. This value can be tuned
to the recommended setting of 3 on T1/E1 (or slower) links using the command shown in
Example 13-2.
Example 13-2 Tuning the Tx-ring
WAG-7206-Left(config)#interface Serial 1/0
WAG-7206-Left(config-if)#tx-ring-limit
tx-ring-limit 3
The new setting quickly can be veried with the same show controllers command, as
Example 13-3 Verifying Tx-ring Changes
WAG-7206-Left#show controllers ser 1/0 | include tx_limited
Tx_underrun_err=0, tx-soft-underru_rr=0, tx-limited=1(3)
3
WAG-7206_Left#
452
NOTE
In ATM, the length of the Tx-ring is dened in (576-byte) particles, not packets, and is
tuned on a per-PVC basis. On some non-ATM interfaces, the Tx-ring even can be tuned to
a minimum of 1 (packet). In either case, the Tx-ring can be tuned (on 768 kbps links) to
approximately 1500 bytes, which is the MTU of Ethernet.
PAK_priority
Chapter 5 introduced PAK_priority, the internal Cisco IOS mechanism for protecting
routing and control trafc. The design implications of PAK_priority are summarized in the
following list:
Layer 2 and Layer 3 control trafc on moderately congested WAN links typically is
protected adequately with the default PAK_priority treatment within the router and
the IP ToS byte markings of IPP6/CS6.
On heavily congested links, it might be necessary to explicitly provision a CBWFQ

bandwidth class for routing/control trafc, as identied by either IPP or CS6.
Although IS-IS trafc receives PAK_priority within the router, it cannot be marked to
IPP6/CS6 because IS-IS uses a CLNS protocol. (It does not use IP, so there are no IPP
or DSCP elds to mark.) This is important to keep in mind if explicit bandwidth
provisioning is required for IS-IS trafc because it cannot be matched against IPP6/
CS6 like most other IGPs. However, NBAR can be used within a class map to match
IS-IS trafc (for example, match protocol clns_is).
Although BGPs (both eBGPs and iBGPs) are marked to IPP6/CS6, they do not receive
PAK_priority treatment within the routers. Therefore, it may be necessary to
provision a separate bandwidth class to protect BGP sessions, even on moderately
congested links where the underlying IGPs are stable.
On Catalyst 6500 switches running Cisco IOS Software on both the supervisors and
MSFC, IGP packets marked internally with PAK_priority additionally are marked
with IPP6/CS6 and the Layer 2 CoS value of 6. This is because scheduling and
congestion avoidance within Cisco Catalyst switches is performed against Layer 2
CoS values.
Link Speeds
In the context of WAN links, there are three main groupings of link speeds. These link
speeds and their respective design implications are summarized in the following list:
Slow (link speed 768 kbps):

Deployment of Interactive-Video generally is not recommended on these
links because of serialization implications.
These links require LFI to be enabled if VoIP is to be deployed over them.
WAN Edge Classication and Provisioning Models
453
cRTP is recommended (with a watchful eye on CPU levels).

Check Tx-ring sizes (especially on slow-speed ATM PVCs); tune to 3, if
needed.
Three- to ve-class trafc models are recommended.
Medium (768 kbps link speed T1/E1):

VoIP or Interactive-Video can be assigned to the LLQ (usually, there is not
enough bandwidth to do both and still keep the LLQ provisioned at less than
33 percentalternatively, Interactive-Video can be placed in a CBWFQ
queue).
LFI is not required.
cRTP is optional.
Three- to ve-class trafc models are recommended.
High ( T1/E1 link speeds):

LFI is not required.
cRTP generally is not recommended (because the cost of increased CPU
levels typically offsets the benets of the amount of bandwidth saved).
Five- to 11-class trafc models are recommended.
Distributed Platform QoS and Consistent QoS Behavior

It is important to keep in mind that minor differences might exist between QoS congurations
on distributed platforms (such as the Cisco 7500 series with VIPs) and those on nondistributed platforms (such as the Cisco 7200 or 1700). The most common difference is the inclusion of the distributed keyword after commands such as ip cef on distributed platforms.
Where more complicated differences exist, they are highlighted explicitly in this chapter.
An important initiative is under way within Cisco to port the QoS code from the Cisco 7500
series routers to the nondistributed router families. This initiative is called Consistent QoS
Behavior and has as its objectives simplifying QoS and increasing QoS consistency between
platforms. Consistent QoS Behavior code should remove most, if not all, conguration
idiosyncrasies between distributed and nondistributed platforms.

One of the most common questions raised when planning a QoS deployment over the WAN
is How many classes of trafc should be provisioned for? The following considerations
should be kept in mind when arriving at an appropriate trafc class model for a given
enterprise.
454
Slow/Medium Link-Speed QoS Class Models

Slow-speed ( 768 kbps) links have very little bandwidth to carve up, to begin with. When
the serialization implications of sending Interactive-Video into the LLQ are taken into
consideration, it becomes generally impractical to deploy more than ve classes of trafc
over slow-speed links.
Medium-speed ( T1/E1) links do not have serialization restrictions and can accommodate
either VoIP or Interactive-Video in their LLQs. However, typically both types of trafc
cannot be provisioned at the same time without oversubscribing the LLQ (provisioning
more than 33 percent of the trafc for the LLQ). Although this might be possible to
congure (the parser will accept the policy and attach it to the interface), the administrator
should remember the trade-off of signicantly adverse data application response times
when LLQs exceed one-third of the link. An alternative approach might be to provision
Interactive-Video in a CBWFQ on medium-speed links.
Three-Class (Voice and Data) Model

If the business objective is simply to deploy VoIP over the existing data network, the Voice
and Data WAN Edge Model is appropriate. Although it might seem that this is a two-class
model, it is actually three: Voice, Call-Signaling, and (generic) data.
Voice is identied by DSCP EF, which is set by default on Cisco IP phones. When identied, VoIP is admitted into the LLQ, which, in this example, is set to the maximum recommended value of 33 percent of the link. Call admission control (CAC) correspondingly
should be assigned to this link by dividing the allocated bandwidth by the voice codec
(including Layer 2 overhead) to determine how many calls can be permitted simultaneously
over this link. Because class-based cRTP is used in this example to compress voice trafc,
it also should be factored into the CAC calculation.
Call-Signaling trafc also is marked on the IP phones (to AF31 currently, but it will be
migrated to CS3, per the QoS Baseline) and requires a relatively small but dedicated
bandwidth guarantee. All other data is fair-queued within class-default. This Three-class
WAN Edge Model is illustrated in Figure 13-2 and detailed in Example 13-4.
455
Figure 13-2 Three-Class WAN Edge Model Migration Strategy Example

Voice
33%
Best-Effort
(62%)
Call-Signaling
5%
Example 13-4 Three-Class WAN Edge Model

!
class-map
match ip
class-map
match ip
match ip
match-all Voice
dscp ef
! IP Phones mark Voice to EF
match-any Call Signaling
dscp cs3
! Future Call-Signaling marking
dscp af31
! IP Phones mark Call-Signaling to AF31
!
policy-map WAN-EDGE
class Voice
priority percent 33
compress header ip rtp
class Call Signaling
bandwidth percent 5
class class-default
fair-queue
!
NOTE
! Maximum recommended LLQ value

! Optional: Enables Class-Based cRTP
! BW guarantee for Call-Signaling
! All other data gets fair-queuing
Sometimes administrators explicitly create a class map that functions as the MQC classdefault. For instance, an administrator might create a class along the lines of that shown in
the following code:
match any
456
or even:
...
access-list 101 permit ip any any
These additional congurations are superuous and inefcient for the router to process. The
MQC implicit class-default should be used instead.
Another advantage of using the MQC implicit class-default is that (currently, before
Consistent QoS Behavior code) on nondistributed platforms, class-default is the only class
that supports fair queuing within it.
Verication command:
show policy
Verication Command: show policy

The preceding three-class policy, like any other MQC policy, can be veried using the show
policy command, as shown in Example 13-5.
Example 13-5 Verication of Three-Class WAN Edge Policy
RBR-2691-Right#show policy WAN-EDGE
Policy Map WAN-EDGE
Class VOICE
! Voice will get LLQ
Strict Priority
! LLQ is provisioned to 33%
Bandwidth 33 (%)
compress:
header ip rtp
! cRTP is enabled
Class CALL-SIGNALING
Bandwidth 5 (%) Max Threshold 64 (packets) ! Call-Signaling gets 5% BW
Class class-default
Flow based Fair Queueing
! Data will get FQ
Bandwidth 0 (kbps) Max Threshold 64 (packets)
RBR-2691-Right#
The Five-Class WAN Edge Model builds on the previous Three-Class WAN Edge Model
and includes a provision for a Critical Data class and a Scavenger class.
The new Critical Data class requires Transactional Data trafc to be marked to DSCP AF21
(or AF22, in the case of dual-rate policers deployed within the campus). Additionally, IGP
routing (marked by the routers as CS6) and Network-Management trafc (recommended to
be marked to CS2) are protected within this class. In this example, the Critical Data class
is provisioned to 36 percent of the link and DSCP-based WRED is enabled on it.
457
The Scavenger class constrains any trafc marked to DSCP CS1 to 1 percent of the link;
this allows class-default to use the remaining 25 percent. However, to constrain Scavenger
to 1 percent, an explicit bandwidth guarantee (of 25 percent) must be given to the BestEffort class. Otherwise, if class-default is not explicitly assigned a minimum bandwidth
guarantee, the Scavenger class still can rob it of bandwidth. This is because of the way the
CBWFQ algorithm has been coded: If classes protected with a bandwidth statement are
offered more trafc than their minimum bandwidth guarantee, the algorithm tries to protect
such excess trafc at the direct expense of robbing bandwidth from class-default (if classdefault is congured with fair-queue), unless class-default itself has a bandwidth
statement (providing itself with a minimum bandwidth guarantee). However, assigning a
bandwidth statement to class-default (on nondistributed platforms) currently precludes the
enabling of fair queuing (fair-queue) on this class and forces FIFO queuing on classdefault (this limitation is to be removed with the release of Consistent QoS Behavior code).
NOTE
An additional implication of using a bandwidth statement on class-default is that even

though 25 percent of the link is reserved explicitly for class-default, the parser will not
attach the policy to an interface unless the max-reserved-bandwidth 100 command is
entered on the interface before the service-policy output statement. This is because the
parser adds the sum of the bandwidth statements (regardless of whether one of these is
applied to the class-default) and, if the total is in excess of 75 percent of the links
bandwidth, rejects the application of the policy to the interface. This is shown in the
following code:
!
description T1 to Branch#60
ip address 10.1.112.1 255.255.255.252
max-reserved-bandwidth 100
! overrides the default 75% BW limit
service-policy output WAN-EDGE
! attaches the MQC policy
ppp multilink
!
Furthermore, WRED can be enabled on the Best-Effort class to provide congestion

management. Because all trafc assigned to the default class is to be marked to the same
DSCP value (of 0), it would be superuous to enable DSCP-based WRED on such a class;
WRED (technically, RED, in this case because all the [IP Precedence] weights are the
same) would sufce.
This Five-Class WAN Edge Model is illustrated in Figure 13-3 and detailed in Example 13-6.
458
Figure 13-3 Five-Class WAN Edge Model Bandwidth Allocation Example

Voice
33%
Best-Effort
25%
Scavenger 1%
Call-Signaling
5%
Critical Data
36%
Example 13-6 Five-Class WAN Edge Model

!
class-map match-all Voice
match ip dscp ef
class-map match-any Call Signaling
match ip dscp cs3
bandwidth percent 1
class-map match-any Critical Data
match ip dscp cs6
match ip dscp cs2
class-map match-all Scavenger
match ip dscp cs1
!
policy-map WAN-EDGE
class Voice
priority percent 33
bandwidth percent 5
class Critical Data
class Scavenger
bandwidth percent 1
class class-default
random-detect
!

! Current Call-Signaling marking
! Routers mark Routing traffic to CS6
! Recommended markings for Transactional-Data
! Recommended marking for Network Management
! Scavenger marking
! Voice gets 33% of LLQ

! Critical Data class gets 36% BW guarantee
! Enables DSCP-WRED for Critical-Data class
! Scavenger class is throttled
! Default class gets a 25% BW guarantee
! Enables WRED for class-default
459
Verication command:
show policy
High Link Speed QoS Class Models

High-speed links (such as multiple T1/E1 or above speeds) allow for the provisioning
of Voice, Interactive-Video, and multiple classes of data, according to the design rules
presented in this chapter (for example, 25 percent for Best Effort class and 33 percent
for all LLQs).
Enabling QoS only optimizes the efciency of bandwidth utilization; it does not create
bandwidth. Therefore, it is important to have adequate bandwidth for all the applications
being provisioned. Furthermore, as WAN bandwidth is becoming less expensive, higherspeed links are becoming more popular.
Even if adequate bandwidth exists for up to 11 classes of trafc, as outlined by the QoS
Baseline Model, not all enterprises are comfortable with deploying such complex QoS
policies at this time. Therefore, it is recommended to start simple, but with room to grow
into more complex models. Figure 13-4 illustrates a simple migration strategy showing
which classes are good candidates for subdivision into more granular classes as future
needs arise.
Figure 13-4 Number of QoS Classes Migration Strategy Example
Five-Class Model
Eight-Class Model
Voice
Voice
Video
Streaming-Video
Call-Signaling
Call-Signaling
Real-Time
Call-Signaling
QoS Baseline Model
Interactive-Video
IP Routing
Network-Control
Critical Data
Critical Data
Network-Management
Transactional Data
Bulk Data
Bulk Data
Best-Effort
Best-Effort
Best-Effort
Scavenger
Scavenger
Scavenger
460
If the enterprises QoS requirements exceed that which the Five-Class Model can provision
for (such as requiring service guarantees for Interactive-Video and requiring Bulk Data to
be controlled during busy periods), they might consider migrating to the Eight-Class Model.
Eight-Class Model
The Eight-Class Model introduces a dual-LLQ design: one for Voice and another for
Interactive-Video.
As pointed out in Chapter 5, the LLQ has an implicit policer that allows for time-division
multiplexing of the single priority queue. This implicit policer abstracts the fact that there
is essentially a single LLQ within the algorithm and, thus, allows for the provisioning of
multiple LLQs.
Interactive-video (or IP videoconferencing, known also as IP/VC) is recommended to be
marked AF41 (which can be marked down to AF42 in the case of dual-rate policing at the
campus access edge). It is recommended to overprovision the LLQ by 20 percent of the
IP/VC rate. This takes into account IP/UDP/RTP headers as well as Layer 2 overhead.
Additionally, Cisco IOS Software automatically includes a 200-ms burst parameter (dened
in bytes) as part of the priority command. On dual-T1 links, this has proven sufcient for
protecting a single 384-kbps IP/VC stream; on higher-speed links (such as triple T1s), the
default burst parameter has shown to be insufcient for protecting multiple IP/VC streams.
However, multiple-stream IP/VC quality tested well with the burst set to 30,000 bytes (for
example, priority 920 30000). Our testing did not arrive at a clean formula for predicting
the required size of the burst parameters as IP/VC streams continually were added; however, given the variable packet sizes and rates of these Interactive-Video streams, this is not
surprising. The main point is that the default LLQ burst parameter might require tuning as
multiple IP/VC streams are added (which likely will be a trial-and-error process).
Optionally, DSCP-based WRED can be enabled on the Interactive-Video class, but testing
has shown negligible performance difference in doing so (because, as already has been
noted, WRED is more effective on TCP-based ows than UDP-based ows, such as
Interactive-Video).
In these designs, WRED is not enabled on classes such as Call-Signaling, IP Routing, or
Network-Management because WRED would take effect only if such classes were lling
their queues nearly to their limits. Such conditions would indicate a provisioning problem
that would better be addressed by increasing the minimum bandwidth allocation for the
class than by enabling WRED.
Additionally, the Eight-Class Model subdivides the preferential data class to separate
control plane trafc (IP routing and Network-Management applications) from businesscritical data trafc. Interior Gateway Protocol (such as RIP, EIGRP, OSPF, and IS-IS)
packets are protected through the PAK_priority mechanism within the router. However,
461
EGP protocols, such as BGP, do not get PAK_priority treatment and might need explicit
bandwidth guarantees to ensure that peering sessions do not reset during periods of
congestion. Additionally, administrators might want to protect network-management
access to devices during periods of congestion.
The other class added to this model is for bulk trafc (Bulk Data class), which is also spun
away from the Critical Data class. Because TCP continually increases its window sizes,
which is especially noticeable in long sessions (such as large le transfers), constraining
Bulk Data to its own class alleviates other data classes from being dominated by such large
le transfers. Bulk Data is identied by DSCP AF11 (or AF12, in the case of dual-rate
policing at the campus access edges). DSCP-based WRED can be enabled on the Bulk Data
class (and also on the Critical Data class).
Figure 13-5 shows sample bandwidth allocations of an Eight-Class Model (for a dual-T1
link example). Figure 13-5 also shows how this model can be derived from the Five-Class
Model in a manner that maintains respective bandwidth allocations as consistently as
possible, which increases the overall end-user transparency of such a migration.
Figure 13-5 Eight-Class WAN Edge Model Bandwidth Allocations Example
Voice
18%
Best-Effort
25%
Interactive-Video
15%
Scavenger 1%
Bulk Data
4%
Critical Data
27%
Call-Signaling
5%
Network-Control
5%
*Dual-T1 Link Example
Example 13-7 shows the corresponding conguration (over a dual-T1 link) for the EightClass Model.
462
Example 13-7 Eight-Class WAN Edge Model

!
match ip dscp ef
class-map match-all Interactive Video
! Recommended markings for IP/VC
match ip dscp cs3
match ip dscp af31
class-map match-any Network Control
match ip dscp cs6
match ip dscp cs2
class-map match-all Critical Data
class-map match-all Bulk Data
! Recommended markings for Bulk-Data
match ip dscp cs1
! Scavenger marking
!
policy-map WAN-EDGE
class Voice
priority percent 18
! Voice gets 552 kbps of LLQ
class Interactive Video
! 384 kbps IP/VC needs 460 kbps of LLQ
priority percent 15
bandwidth percent 5
class Network Control
! Routing and Network Management get min 5% BW
bandwidth percent 5
class Critical Data
! Critical Data gets min 27% BW
! Enables DSCP-WRED for Critical-Data class
class Bulk Data
! Bulk Data gets min 4% BW guarantee
bandwidth percent 4
! Enables DSCP-WRED for Bulk-Data class
class Scavenger
bandwidth percent 1
class class-default
! Fair-queuing is sacrificed for BW guarantee
random-detect
! Enables WRED on class-default
!
!
NOTE
463
The Consistent QoS Behavior initiative will enable the conguration of a bandwidth
statement along with fair-queue on any class, including class-default, on all platforms.
Verication command:
show policy
QoS Baseline (11-Class) Model

As mentioned in the overview, the QoS Baseline is a guiding model for addressing the QoS
needs of today and the foreseeable future. The QoS Baseline is not a mandate dictating what
enterprises must deploy today; instead, this strategic document offers standards-based
recommendations for marking and provisioning trafc classes that will allow for greater
interoperability and simplied future expansion.
Building on the previous model, the Network-Control class is subdivided into the IP
Routing and Network-Management classes.
The Critical Data class also is subdivided further into the Mission-Critical Data and
Transactional Data classes. Although DSCP-based WRED is enabled on the Transactional
Data class, because packets for this class can be marked AF21 (or AF22, as in the case of
dual-rate policers being deployed in the campus), it would be superuous to enable DSCPbased WRED on the Mission-Critical Data class (WRED will sufce because all MissionCritical Data class packets are marked to the same value: DSCP 25).
Finally, a new class is provisioned for Streaming-Video. Testing has shown that there is a
negligible difference in enabling WRED on this UDP-based trafc class, so, although it
remains an option, WRED is not enabled in these design examples.
Figure 13-6 shows a sample WAN edge bandwidth allocation for a QoS Baseline Model
(over a dual-T1 link) and also shows how this model can be derived from the Five- and
Seven-Class Models in a manner that maintains respective bandwidth allocations as
consistently as possible. This increases the overall end-user transparency of such a
migration.
464
Figure 13-6 QoS Baseline WAN Edge Model Bandwidth Allocations Example
Voice
18%
BestEffort
25%
InteractiveVideo
15%
Scavenger 1%
Bulk 4%
Streaming-Video
10%
Call-Signaling 5%
Routing 3%
Network-Management 2%
Transactional Data
7%
Mission-Critical Data 10%

* Dual-T1 Link Example
Example 13-8 shows the corresponding conguration for an 11-Class QoS Baseline WAN
Edge Model (over a dual-T1 link).
Example 13-8 QoS Baseline WAN Edge Model
!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
class-map match-all Routing
match ip dscp cs6
class-map match-all Net Mgmt
match ip dscp cs2
class-map match-all Mission-Critical Data
! Interim marking for Mission-Critical Data
match ip dscp 25
class-map match-all Transactional Data
465
Example 13-8 QoS Baseline WAN Edge Model (Continued)

class-map match-all Streaming
match ip dscp cs4
match ip dscp cs1
!
policy-map WAN-EDGE
class Voice
priority percent 18
priority percent 15
bandwidth percent 5
class Routing
bandwidth percent 3
class Net Mgmt
bandwidth percent 2
class Mission-Critical Data
random-detect
class Transactional Data
bandwidth percent 7
class Bulk Data
bandwidth percent 4
class Streaming Video
class Scavenger
bandwidth percent 1
class class-default
random-detect
!

Video
! Recommended marking for Streaming-Video
! Recommended marking for Scavenger traffic

! Routing class gets explicit BW guarantee
! Net-Mgmt class gets explicit BW guarantee
! Mission-Critical class gets 10% BW guarantee
! Enables WRED for Mission-Critical Data class
! Transactional-Data class gets 7% BW guarantee
! Enables DSCP-WRED for Transactional-Data class
! Bulk Data remains at 4% BW guarantee
! Enables DSCP-WRED for Bulk-Data class
! Streaming-Video class gets 10% BW guarantee
! Class-Default gets 25% min BW guarantee
! Enables WRED on class-default
Verication command:
show policy
Again, a bandwidth statement is used on class-default (currently), precluding the use of

fair-queue on the class for all nondistributed platforms. Also, a max-reserved-bandwidth
100 statement must be applied to the interface before the service-policy output statement.
Distributed-Platform/Consistent QoS BehaviorQoS Baseline Model

One of the current advantages of the Cisco 7500 (distributed platform) QoS code is that it
can support bandwidth commands in conjunction with fair-queue on any given class,
including class-default. This functionality will become available to nondistributed platforms
with the release of Consistent QoS Behavior code. (As of this writing, this initiative does
466
not have a xed target delivery date.) When fair-queue is enabled on the main data classes,
the resulting conguration becomes as shown in Example 13-9.
Example 13-9 Distributed-Platform/Consistent QoS BehaviorQoS Baseline WAN Edge Model
!
ip cef distributed
! 'distributed' keyword required on 7500 for ip cef
!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
match ip dscp cs6
match ip dscp cs2
match ip dscp 25
class-map match-all Streaming Video
match ip dscp cs4
match ip dscp cs1
!
policy-map WAN-EDGE
class Voice
priority percent 18
priority percent 15
bandwidth percent 5
! Bandwidth guarantee for Call-Signaling
class Routing
bandwidth percent 3
! Bandwidth guarantee for Routing
class Net Mgmt
bandwidth percent 2
! Bandwidth guarantee for Network Management
! Mission-Critical data gets min 10% BW guarantee
fair-queue
! Applies FQ to Mission-Critical Data class
random-detect
! Enables WRED on Mission-Critical Data class
bandwidth percent 7
! Transactional Data gets min 7% BW guarantee
fair-queue
! Applies FQ to Transactional Data class
! Enables DSCP-WRED on Transactional Data class
class Bulk Data
bandwidth percent 4
fair-queue
! Applies FQ to Bulk Data class
WAN Edge Link-Specic QoS Design
467
Example 13-9 Distributed-Platform/Consistent QoS BehaviorQoS Baseline WAN Edge Model (Continued)
class Scavenger
bandwidth percent 1
class class-default
fair-queue
random-detect
! Enables DSCP-WRED on Bulk Data class

! Streaming-Video gets min 10% BW guarantee
! Class-Default gets min 25% BW guarantee
! Applies FQ to Class-Default
! Enables WRED on Class-Default

The most popular WAN media in use today are leased lines, Frame Relay, and ATM
(including ATM-to-Frame Relay Service Interworking). Each of these media can be
deployed in three broad categories of link speeds: slow speed ( 768 kbps), medium speed
( T1/E1), and high speed (multiple T1/E1 or greater). The following sections detail
specic designs for each medium at each speed category. Additionally, ISDN QoS design
is discussed in the context of a backup WAN link.
Leased Lines
Leased lines, or point-to-point links, can be congured with HDLC, PPP, or MLP encapsulation. MLP offers the network administrator the most exibility and deployment options.
For example, MLP is the only leased-line protocol that supports LFI on slow-speed links
(through MLP LFI). Additionally, as bandwidth requirements grow over time, MLP requires
the fewest modications to accommodate the addition of multiple T1/E1 lines to a WAN
link bundle. Furthermore, MLP supports all of the security options of PPP (such as CHAP
authentication).
Slow-Speed ( 768 kbps) Leased Lines

Recommendation: Use MLP LFI and cRTP.
For slow-speed leased lines (as illustrated in Figure 13-7), LFI is required to minimize
serialization delay. MLP, therefore, is the only encapsulation option on slow-speed leased
lines because MLP LFI is the only mechanism available for fragmentation and interleaving
on such links. Optionally, cRTP can be enabled either as part of the MQC policy map
(as shown in Example 13-10) or under the multilink interface (using the ip rtp headercompression command). Ensure that MLP LFI and cRTP, if enabled, are congured on
both ends of the point-to-point link, as shown in Example 13-14.
468
Figure 13-7 Slow-Speed Leased Lines

MLP Link
768 kbps
Branch Router
WAN Aggregator
Example 13-10 Slow-Speed ( 768 kbps) Leased-Line QoS Design Example

!
policy-map WAN-EDGE
class Voice
priority percent 33
! Maximum recommended LLQ value
! Enables Class-Based cRTP
bandwidth percent 5
! A 3 to 5 Class Model can be used

!
description 768 kbps Leased-Line to RBR-3745-Left
ip address 10.1.112.1 255.255.255.252
service-policy output WAN-EDGE ! Attaches the MQC policy to Mu1
ppp multilink
ppp multilink fragment delay 10 ! Limits serialization delay to 10 ms
! Enables interleaving of Voice with Data
!
!
interface Serial1/0
bandwidth 786
no ip address
encapsulation ppp
ppp multilink
! Includes interface Ser1/0 into Mu1 group
!
Verication commands:
show policy
show interface
show ppp multilink
469
Verication Command: show interface

The show interface command indicates whether drops are occurring on an interface (an
indication of congestion). Additionally, on a multilink interface with LFI enabled, the
command displays interleaving statistics, as shown in Example 13-11.
Example 13-11 show interface Verication of MLP LFI on a Slow-Speed Leased Line
WAG-7206-Left#show interface multilink 1
Multilink1 is up, line protocol is up
Hardware is multilink group interface
Description: 768 kbps Leased-Line to RBR-3745-Left
Internet address is 10.1.112.1/30
MTU 1500 bytes, BW 768 Kbit, DLY 100000 usec,
reliability 255/255, txload 233/255, rxload 1/255
Encapsulation PPP, LCP Open, multilink Open
Open: CDPCP, IPCP, loopback not set
DTR is pulsed for 2 seconds on reset
Last input 00:00:01, output never, output hang never
Last clearing of "show interface" counters 00:16:15
Input queue: 0/75/0/0 (size/max/drops/flushes);
Total output drops: 49127
Queueing strategy: weighted fair
Output queue: 54/1000/64/49127/185507
49127 185507
(size/max total/threshold/drops/interleaves)
drops interleaves
In Example 13-11, 49,127 drops have occurred on the multilink interface (because of
congestion), and LFI has engaged with 185,507 interleaves of voice with data.
Verication Command: show policy interface (Three-Class Policy)

The show policy interface command is probably the most useful show command for
MQC-based QoS policies. It displays a wide array of dynamic statistics, including the
number of matches on a class map as a whole, the number of matches against each discrete
match statement within a class map, the number of queued or dropped packets (either tail
dropped or WRED dropped), and many other relevant QoS statistics. Example 13-12 shows
example output of the show policy interface command.
Example 13-12 show policy interface Verication of a Three-Class Policy on a Slow-Speed Leased
Line
WAG-7206-Left#show policy interface multilink 1
Multilink1
Service-policy output: WAN-EDGE
Class-map: Voice (match-all)
Match: ip dscp ef
Queueing
Strict Priority
continues
470
Example 13-12 show policy interface Verication of a Three-Class Policy on a Slow-Speed Leased
Line (Continued)
Output Queue: Conversation 264
Bandwidth 33 (%)
Bandwidth 253 (kbps) Burst 6325 (Bytes)
(pkts matched/bytes matched) 68392/2043848
(total drops/bytes drops) 0/0
compress:
header ip rtp
UDP/RTP compression:
Sent: 68392 total, 68388 compressed,
2333240 bytes saved, 1770280 bytes sent
2.31 efficiency improvement factor
99% hit ratio, five minute miss rate 0 misses/sec,0 max
rate 41000 bps
Class-map: Call Signaling (match-any)
Match: ip dscp cs3
0 packets, 0 bytes
30 second rate 0 bps
Match: ip dscp af31
Queueing
Bandwidth 5 (%)
(depth/total drops/no-buffer drops) 0/0/0
Match: any
Queueing
Flow Based Fair Queueing
Maximum Number of Hashed Queues 256
(total queued/total drops/no-buffer drops) 36/458/0
WAG-7206-Left#
In Example 13-12, the Voice class map and Call-Signaling class map are receiving matches
on their classication criteria (DSCP EF and DSCP CS3/AF31, respectively). However,
because Cisco IP Telephony products currently mark Call-Signaling trafc to DSCP AF31,
Call-Signaling trafc is matching only on DSCP AF31 in this example.
The last line of every class map output is important because this line indicates whether any
drops are occurring on this trafc class. In this example, there are no drops in the Voice or
Call-Signaling classes, which is the desired behavior. A few drops are occurring in classdefault, but this is expected when the interface is congested (which is the trigger to engage
queuing).
471
Also of note, and specic to this particular conguration, are the cRTP statistics included
under the Voice class map. These cRTP statistics are displayed because class-based cRTP
was enabled in this example (instead of enabling cRTP on the interface). Remember, cRTP
must be enabled on both ends of the links for compression to occur; otherwise, these
counters will never increment.
Medium-Speed ( T1/E1) Leased Lines

Recommendation: MLP LFI is not required; cRTP is optional.
Medium-speed leased lines (as shown in Figure 13-8) can use HDLC, PPP, or MLP
encapsulation. An advantage of using MLP encapsulation is that future growth (to multiple
T1/E1 links) will be easier to manage. Also, MLP includes all the security options of PPP
(such as CHAP).
Figure 13-8 Medium-Speed Leased Lines
MLP T1/E1 Link
Branch Router
WAN Aggregator
However, MLP LFI is not required at these speeds, and cRTP is optional. Example 13-13
shows an example conguration for medium-speed leased lines.
Example 13-13 Medium-Speed Leased-Line QoS Design Example
!
description T1 Leased-Line to RBR-3745-Left
ip address 10.1.112.1 255.255.255.252
! Attaches the MQC policy to Mu1
ppp multilink
! Identifies Mu1 as logical Int for Mu1 group
!
!
interface Serial1/0
bandwidth 1536
no ip address
encapsulation ppp
load-interval 30
ppp multilink
! Includes interface Ser1/0 into Mu1 group
!
472
show policy
show interface
High-Speed (Multiple T1/E1 or Greater) Leased Lines

Recommendation: Use MLP bundling, but keep an eye on CPU levels. When enterprises
have multiple T1/E1-speed leased lines to individual branches, three options exist for load
sharing:
IP CEF per-destination load balancing

IP CEF per-packet load balancing
Multilink PPP bundles
Cisco Technical Marketing testing has shown that IP CEF per-destination load balancing
does not meet the SLAs required for Voice and Interactive-Video over multiple T1/E1 links,
as shown in Figure 13-9.
Figure 13-9 High-Speed Leased Lines
MLP Link
Multiple T1/E1
Branch Router
WAN Aggregator
On the other hand, IP-CEF per-packet load balancing did meet the required SLAs, but not
quite as well as MLP bundling.
MLP bundling attained the best overall SLA values for delay and jitter, but it required more
CPU resources than IP CEF per-packet load balancing. If CPU levels are kept under the
recommended 75 percent, it is recommended to use MLP bundling for multiple T1/E1
links.
Also, if policy maps that require bandwidth statements on class-default are being attached
to the multilink interface, the max-reserved-bandwidth 100 command is required on
the interface before the service-policy output statement can be applied, as shown in
Example 13-14.
473
Example 13-14 High-Speed ( Multiple T1/E1) Leased Line QoS Design Example
!
description Dual-T1 to RBR-3745-Left
ip address 10.1.112.1 255.255.255.252
! Overrides the default 75% BW limit
! Attaches the MQC policy to Mu1
ppp multilink
! Identifies Mu1 as logical int for Mu1 group
!
!
interface Serial1/0
! defined on physical interface only
bandwidth 1536
no ip address
encapsulation ppp
ppp multilink
! includes interface Ser1/0 into Mu1 group
!
interface Serial1/1
! defined on physical interface only
bandwidth 1536
no ip address
encapsulation ppp
ppp multilink
! includes interface Ser1/1 into Mu1 group
!
NOTE
Interface bandwidth commands (not to be confused with policy map CBWFQ bandwidth
commands) should be dened only on the physical interfaces, not on multilink interfaces.
This way, if any physical interfaces go down, the Cisco IOS Software will reect the change
in the multilink interfaces bandwidth for routing and QoS purposes. This change can be
veried by the show interface command. However, if a bandwidth statement is congured
under the multilink interface, the bandwidth value for the interface will be static even if an
underlying physical interface is lost.
show policy
show interface
show ppp multilink
474
Verication Command: show policy interface (QoS Baseline Policy)

A more complex example of the show policy interface command is given in Example
13-15, where a QoS Baseline WAN edge policy is being applied to a dual-T1 (high-speed)
leased line.
Example 13-15 show policy interface Verication of a QoS Baseline Policy on a High-Speed Leased
Line
WAG-7206-Left#show policy interface multilink 1
Multilink1
Service-policy output: WAN-EDGE
Class-map: Voice (match-all)
Match: ip dscp ef
Queueing
Strict Priority
Bandwidth 18 (%)
Class-map: Interactive Video (match-all)
Match: ip dscp af41
Queueing
Strict Priority
Bandwidth 15 (%)
Class-map: Call Signaling (match-any)
Match: ip dscp cs3
0 packets, 0 bytes
Match: ip dscp af31
Queueing
Bandwidth 5 (%)
Class-map: Routing (match-all)
Match: ip dscp cs6
Queueing
475
Line (Continued)
Bandwidth 3 (%)
Class-map: Net Mgmt (match-all)
Match: ip dscp cs2
Queueing
Bandwidth 2 (%)
Class-map: Mission-Critical Data (match-all)
Match: ip dscp 25
Queueing
Bandwidth 12 (%)
Bandwidth 368 (kbps)
exponential weight: 9
mean queue depth: 10
class
Transmitted
Random drop
Tail drop
Minimum Maximum
pkts/bytes
pkts/bytes
pkts/bytes thresh thresh
0
0/0
0/0
0/0
20
40
1
0/0
0/0
0/0
22
40
2
0/0
0/0
0/0
24
40
3
57178/41112815
0/0
0/0
26
40
4
0/0
0/0
0/0
28
40
5
0/0
0/0
0/0
30
40
6
0/0
0/0
0/0
32
40
7
0/0
0/0
0/0
34
40
rsvp
0/0
0/0
0/0
36
40
Class-map: Transactional Data (match-all)
Match: ip dscp af21
Queueing
Bandwidth 8 (%)
mean queue depth: 26
Mark
prob
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
continues
476
Line (Continued)
class
Transmitted
Random drop
Tail drop
Minimum Maximum
pkts/bytes
pkts/bytes
0
0/0
0/0
0/0
20
40
1
0/0
0/0
0/0
22
40
2
30787/31019741
954/988392
0/0
24
40
3
0/0
0/0
0/0
26
40
4
0/0
0/0
0/0
28
40
5
0/0
0/0
0/0
30
40
6
0/0
0/0
0/0
32
40
7
0/0
0/0
0/0
34
40
rsvp
0/0
0/0
0/0
36
40
Class-map: Streaming Video (match-all)
Match: ip dscp cs4
Queueing
Bandwidth 10 (%)
Class-map: Scavenger (match-all)
285075 packets, 129433625 bytes
Match: ip dscp cs1
Queueing
Bandwidth 1 (%)
Match: any
Queueing
Bandwidth 25 (%)
mean queue depth: 4
class
Transmitted
Random drop
Tail drop
Minimum Maximum
pkts/bytes
pkts/bytes
0
40961/35700528
268/217632
0/0
20
40
1
0/0
0/0
0/0
22
40
2
0/0
0/0
0/0
24
40
Mark
prob
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
Mark
prob
1/10
1/10
1/10
477
Line (Continued)
3
4
5
6
7
rsvp
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
26
28
30
32
34
36
40
40
40
40
40
40
1/10
1/10
1/10
1/10
1/10
1/10
Important items to note for a given class are the pkts matched statistics (which verify that
classication has been congured correctly and that the packets have been assigned to the
proper queue) and the total drops statistics (which indicate whether adequate bandwidth
has been assigned to the class).
Extremely few drops, if any, are desired in the Voice, Interactive-Video, Call-Signaling, and
Routing classes.
NOTE
The Routing class is a special case because of the statistics that it displays.
On nondistributed platforms, the classication counter (the rst line under the class
map) shows any IGP trafc matched by the Routing class (identied by DSCP CS6).
But remember that IGP protocols queue separately (because these are handled by the
PAK_priority mechanism) and, therefore, do not register queuing statistics within the MQC
counters for the Routing class. EGP protocols (such as BGP), on the other hand, do register
queuing/dropping statistics within such an MQC class.
The situation is different on distributed platforms, where all routing packets (IGP or EGP)
are matched and queued within a provisioned Routing class (complete with queuing/
dropping statistics through the show policy interface verication command).
Few drops are expected in the Mission-Critical Data class. WRED (essentially RED
because all packets are marked to the same IPP/DSCP value) is enabled to avoid congestion
on this class. Some drops are expected for the Transactional Data class, yet, in this
particular example, WRED is minimizing tail drops for this class.
It is normal for the Bulk Data class to show drops (both WRED and tail). This is because
the Bulk Data class is being constrained from dominating bandwidth by its large and
sustained TCP sessions. The Scavenger class should show very aggressive dropping during
periods of congestion. Finally, it is normal for drops to appear in the default class.
478
Verication Command: show ppp multilink

The show ppp multilink command is useful to verify that multiple physical links are
correctly associated and included in the MLP bundle, as shown in Example 13-16. Also,
the load (which might not quite hit 255/255) indicates congestion on the link.
Example 13-16 show ppp multilink Verication of a High-Speed Leased Line
WAG-7206-Left#show ppp multilink
Multilink1, bundle name is RBR-3745-Left
Bundle up for 00:28:33, 254/255 load
Receive buffer limit 24384 bytes, frag timeout 1000 ms
0/0 fragments/bytes in reassembly list
0 lost fragments, 2 reordered
0/0 discarded fragments/bytes, 0 lost received
0xE8F received sequence, 0x9A554 sent sequence
Member links: 2 active,
active 0 inactive (max not set, min not set)
Se1/0,
Se1/0 since 00:28:35, 1920 weight, 1496 frag size
Se1/1,
Se1/1 since 00:28:33, 1920 weight, 1496 frag size
Frame Relay
Recommendation: For the latest feature combinations and management options, use classbased Frame Relay trafc shaping whenever possible.
Frame Relay networks are the most popular WANs in use today because of the low costs
associated with them. Frame Relay is a nonbroadcast multiaccess (NBMA) technology that
frequently utilizes oversubscription to achieve cost savings (similar to airlines overselling
seats on ights to achieve maximum capacity and protability).
To manage oversubscription and potential speed mismatches between senders and receivers,
a trafc-shaping mechanism must be used with Frame Relay. Either Frame Relay trafc
shaping (FRTS) or class-based FRTS can be used. The primary advantage of using classbased FRTS is management because shaping statistics and queuing statistics are displayed
jointly with the show policy interface verication command and are included in the
SNMPv2 Cisco class-based QoS Management Information Base (MIB).
FRTS and class-based FRTS require the following parameters to be dened:
Committed information rate (CIR)

Committed burst rate (Bc)
Excess burst rate (Be)
Minimum CIR
Fragment size (required only on slow-speed links)
479
Committed Information Rate

Recommendation: Set the CIR to 95 percent of the PVC contracted speed.
In most Frame Relay networks, a central sites high-speed links connect to lower-speed
links to/from many remote ofces. For example, consider a central site that sends out data
at 1.536 Mbps, while a remote branch might have only a 56-kbps circuit into it. This speed
mismatch can cause congestion delays and drops. In addition, there is typically a many-toone ratio of remote branches to central hubs, making it possible for many remote sites to
send trafc at a rate that can overwhelm the T1 at the hub. Both scenarios can cause frame
buffering in the provider network, which introduces jitter, delay, and loss.
The only solution to guarantee service-level quality is to use trafc shaping at both the
central and remote routers and to dene a consistent CIR at both ends of the Frame Relay
PVC. Because the FRTS mechanism does not take Frame Relay overhead (headers and
cyclic redundancy checks [CRCs]) into account in its calculations, it is recommended that
the CIR be set slightly below the contracted speed of the PVC. Cisco Technical Marketing
testing has shown that setting the CIR to 95 percent of the contracted speed of the PVC
engages the queuing mechanism (LLQ/CBWFQ) slightly early and improves service levels
for Real-Time applications, like Voice.
Committed Burst Rate

Recommendation: Set the Bc to CIR/100 on nondistributed platforms and to CIR/125 on
distributed platforms.
With Frame Relay networks, you also need to consider the amount of data that a node can
transmit at any given time. A 56-kbps PVC can transmit a maximum of 56 kbps of trafc
in 1 second. Trafc is not sent during the entire second, however, but only during a dened
window called the interval (Tc). The amount of trafc that a node can transmit during this
interval is called the committed burst (Bc) rate. By default, Cisco IOS Software sets the Bc
to CIR/8. This formula is used for calculating the Tc follows:
Tc = Bc / CIR
For example, a CIR of 56 kbps is given a default Tc of 125 ms (7000 / 56,000). If the
56-kbps CIR is provisioned on a WAN aggregator that has a T1 line-rate clock speed, every
time the router sends its allocated 7000 bits, it has to wait 120.5 ms before sending the next
batch of trafc. Although this is a good default value for data, it is a bad choice for voice.
By setting the Bc value to a much lower number, you can force the router to send less trafc
per interval, but over more frequent intervals per second. This results in signicant
reduction in shaping delays.
The optimal congured value for Bc is CIR/100, which results in a 10-ms interval
(Tc = B / CIR).
480
On distributed platforms, the Tc must be dened in 4-ms increments. The nearest multiple
of 4 ms within the 10-ms target is 8 ms. This interval can be achieved by conguring the
Bc to equal CIR/125.
Excess Burst Rate

Recommendation: Set the Be to 0.
If the router does not have enough trafc to send all of its Bc (1000 bits, for example), it
can credit its account and send more trafc during a later interval. The maximum amount
that can be credited to the routers trafc account is called the excess burst (Be) rate. The
problem with Be in converged networks is that this can create a potential for buffering
delays within a Frame Relay network (because the receiving side can pull the trafc from
a circuit only at the rate of Bc, not Bc + Be). To remove this potential for buffering delays,
it is recommended to set the Be to 0.
Minimum Committed Information Rate

Recommendation: Set the minCIR to CIR.
The minimum CIR is the transmit value that a Frame Relay router will rate down to when
backward-explicit congestion notications (BECNs) are received. By default, Cisco IOS
Software sets the minimum CIR to CIR/2. However, to maintain consistent service levels,
it is recommended that adaptive shaping be disabled and that the minimum CIR be set equal
to the CIR (which means there is no rating down). An exception to this rule would occur
if a tool such as Frame Relay voice-adaptive trafc shaping was deployed (for more
information and recommendations on using FR-VATS, refer to Chapter 4, Policing and
Shaping Tools).
Slow-Speed ( 768 kbps) Frame Relay Links

Recommendation: Enable FRF.12 and set the fragment size for 10 ms maximum
serialization delay. Enable cRTP.
As with all slow-speed links, slow Frame Relay links (as illustrated in Figure 13-10) require
a mechanism for fragmentation and interleaving. In the Frame Relay environment, the tool
for accomplishing this is FRF.12.
Figure 13-10 Slow-Speed Frame Relay Links
Frame Relay Link
768 kbps
Frame Relay
Cloud
Branch Router
WAN Aggregator
481
Unlike MLP LFI, which takes the maximum serialization delay as a parameter, FRF.12
requires the actual fragment sizes to be dened manually. This requires some additional
calculations because the maximum fragment sizes vary by link speed. These fragment sizes
can be calculated by multiplying the provisioned line-clocking speed by the recommended
maximum serialization delay target (10 ms), and converting the result from bits to bytes
(which is done by dividing the result by 8):
Fragment Size in Bytes = (Link Speed in kbps * Maximum Allowed Jitter in ms) / 8
For example, the calculation for the maximum fragment size for a 56-kbps circuit is as
follows:
Fragment Size = (56 kbps * 10 ms) / 8 = 70 Bytes
Table 13-1 shows the recommended values for FRF.12 fragment sizes, CIR, and Bc for
slow-speed Frame Relay links.
Table 13-1
Recommended Fragment Sizes, CIR, and Bc Values for Slow-Speed Frame Relay Links
PVC Speed
Maximum Fragment
Size (for 10-ms Delay)
Recommended
CIR Values
Recommended
Bc Values
56 kbps
70 bytes
53,200 bps
532 bits per Tc
64 kbps
80 bytes
60,800 bps
608 bits per Tc
128 kbps
160 bytes
121,600 bps
1216 bits per Tc
256 kbps
320 bytes
243,200 bps
2432 bits per Tc
512 kbps
640 bytes
486,400 bps
4864 bits per Tc
768 kbps
960 bytes
729,600 bps
7296 bits per Tc
Both FRTS and class-based FRTS require a Frame Relay map class to be applied to the
DLCI. Also in both cases, the frame-relay fragment command is applied to the map class.
However, unlike FRTS, class-based FRTS does not require frame-relay trafc-shaping to
be enabled on the main interface. This is because MQC-based/class-based FRTS requires a
hierarchal (or nested) QoS policy to accomplish both shaping and queuing. This hierarchical policy is attached to the Frame Relay map class, which is bound to the DLCI.
As with slow-speed leased-line policies, cRTP can be enabled within the MQC queuing
policy under the Voice class. Example 13-17 shows an example of slow-speed Frame Relay
link-specic conguration.
Example 13-17 Slow-Speed ( 768 kbps) Frame Relay QoS Design Example
!
policy-map MQC-FRTS-768
class class-default
service-policy WAN-EDGE
! Enables MQC-Based FRTS

! Queues packets headed to the shaper
continues
482
Example 13-17 Slow-Speed ( 768 kbps) Frame Relay QoS Design Example (Continued)
!
!
interface Serial2/0
no ip address
!
ip address 10.1.121.1 255.255.255.252
! Binds the map-class to the FR DLCI
!
!
! Attaches nested MQC policies to map-class
service-policy output MQC-FRTS-768
! Enables FRF.12
!
show policy map

show policy-map interface
show frame-relay fragment
Verication Command: show frame-relay fragment

The show frame-relay fragment command, shown in Example 13-18, provides verication of the fragment size, regardless of whether regular FRF.12 fragmentation or Frame
Relay voice-adaptive trafc shaping (and fragmentation) is congured for a DLCI. Additionally, dynamic counters monitor how many frames required fragmentation in either
direction.
Example 13-18 show frame-relay fragment Verication of a Slow-Speed Frame Relay Link
WAG-7206-Left#show frame-relay fragment 102
interface
dlci frag-type
frag-size
Serial2/0.12
102
end-to-end
960
WAG-7206-Left#
in-frag
5476
out-frag dropped-frag
2035
0
Medium-Speed ( T1/E1) Frame Relay Links

Recommendation: FRF.12 is not required. cRTP is optional.
The conguration for medium-speed Frame Relay links, illustrated in Figure 13-11 and
detailed in Example 13-19, is identical to that for slow-speed Frame Relay links, with the
exception that enabling FRF.12 no longer is required.
483
Figure 13-11 Medium-Speed Frame Relay Links

Frame Relay Link
T1/E1
Frame Relay
Cloud
Branch Router
WAN Aggregator
NOTE
In some cases, however, administrators have chosen to enable FRF.12 on T1/E1 speed links,
even though the fragment size for a 10-ms maximum serialization delay at such speeds is
greater than the MTU of Ethernet (1500 bytes). The rationale behind doing so is to retain
the Frame Relay dual-FIFO queuing mechanism at Layer 2 (discussed in Chapter 5,
Congestion-Management Tools), which can provide slightly superior service levels under
certain conditions. Generally, this is not required, however.
Example 13-19 Medium-Speed (T1/E1) Frame Relay QoS Design Example

!
class class-default
!
!
interface Serial2/0
no ip address
!
ip address 10.1.121.1 255.255.255.252
!
!
service-policy output MQC-FRTS-1536 ! Attaches nested MQC policies to map-class
!
484
show policy map

High-Speed (Multiple T1/E1 and Greater) Frame Relay Links

Recommendation: Use IP CEF per-packet load balancing for load sharing across multiple
physical Frame Relay links.
When multiple Frame Relay circuits exist between a central WAN aggregation router and
a remote branch router, as illustrated in Figure 13-12, it is recommended that IP CEF perpacket load balancing be used to load-share between the links. Multilink PPP over Frame
Relay (MLPoFR) bundles are complex to congure and difcult to manage, whereas IP
CEF per-packet load balancing is not and has the lowest CPU impact of the load-sharing
mechanisms. Therefore, IP CEF per-packet load balancing is recommended across multiple
Frame Relay links to the same branch.
Figure 13-12 High-Speed Frame Relay Links
Frame Relay Link
Multiple T1/E1
Frame Relay
Cloud
Branch Router
WAN Aggregator
NOTE
It is important to keep in mind that providers might have geographically dispersed paths to
the same sites; therefore, the delay on one T1 FR link might be slightly higher or lower than
the delay on another. This could cause TCP sequencing issues and slightly reduce effective
data application throughput. Network administrators should keep these factors in mind
when planning their WAN topologies.
The max-reserved-bandwidth 100 command is not required on the interfaces because the
queuing policy is not applied directly to the interface; instead, it is applied to another policy
(the MQC-based Frame Relay trafc-shaping policy). Example 13-20 shows the conguration for a high-speed Frame Relay link.
485
Example 13-20 High-Speed ( Multiple T1/E1) Frame Relay QoS Design Example
!
class class-default
!
!
interface Serial2/0
no ip address
no fair-queue
!
ip address 10.1.121.1 255.255.255.252
! Enables IP CEF Per-Packet Load-Sharing
ip load-sharing per-packet
! Binds the map-class to FR DLCI 102
!
interface Serial2/1
no ip address
serial restart_delay 0
!
ip address 10.1.121.5 255.255.255.252
! Enables IP CEF Per-Packet Load-Sharing
ip load-sharing per-packet
! Binds the map-class to FR DLCI 112
!
!
service-policy output MQC-FRTS-1536 ! Attaches nested MQC policies to map-class
!
show policy map

486
Distributed Platform Frame Relay Links

Recommendation: Set CIR values to multiples of 8000. Set the Bc to CIR/125.
When ported to distributed-platform WAN aggregators (such as the Cisco 7500 VIP), most
policies require little more than ensuring that IP CEF is running in distributed mode. However, FRTS is not supported in a distributed environment, so another shaping tool is required.
Distributed trafc shaping (dTS) can be used in conjunction with hierarchical MQC policies
to achieve a similar effect on trafc ows over distributed Frame Relay WAN links. Figure 13-13 shows a Frame Relay link homed from a distributed-platform WAN aggregator.
Figure 13-13 Distributed-Platform Frame Relay Links
Frame Relay Link
Frame Relay
Cloud
Branch Router
Distributed Platform
WAN Aggregator
Although dTS on the Cisco 7500 is very similar to MQC-based FRTS on nondistributed
platforms, there are two main caveats to keep in mind. The rst is that the CIR must be
dened in multiples of 8000. Therefore, it is recommended that the CIR be dened as 95
percent of the PVC speed, rounded down to the nearest multiple of 8000. The second caveat
is that the Cisco 7500 VIP requires the Tc to be dened in an increment of 4 ms. Because
the target interval for all platforms is 10 ms, which is not evenly divisible by 4 ms, the
recommendation for the Cisco 7500 VIP is to use an interval of 8 ms. The interval can
be set to 8 ms by dening the burst using the following formula:
Bc = CIR/125
Table 13-2 gives recommended values for fragment sizes, CIR, and Bc for distributed
platforms. (Some values have been slightly rounded for conguration and management
simplicity.)
Table 13-2
Recommended Fragment Sizes, CIR, and Bc Values for Distributed Platform Frame Relay
Links
PVC Speed
Maximum Fragment
Recommended
CIR Values
Recommended
Bc Values
56 kbps
70 bytes
48,000 bps
384 bits per Tc
64 kbps
80 bytes
56,000 bps
448 bits per Tc
128 kbps
160 bytes
120,000 bps
960 bits per Tc
256 kbps
320 bytes
240,000 bps
1920 bits per Tc
Table 13-2
487
Recommended Fragment Sizes, CIR, and Bc Values for Distributed Platform Frame Relay
Links (Continued)
PVC Speed
Maximum Fragment
Recommended
CIR Values
Recommended
Bc Values
512 kbps
640 bytes
480,000 bps
3840 bits per Tc
768 kbps
960 bytes
720,000 bps
5760 bits per Tc
1536 kbps
1,440,000 bps
11520 bits per Tc
2048 kbps
1,920,000 bps
15360 bits per Tc
Example 13-21 Distributed Platform Frame Relay QoS Design (Slow-Speed Link) Example
!
!
ip cef distributed
! 'distributed' keyword required on 7500 for ip cef
!
!
policy-map MQC-DTS-768
class class-default
! Enables Distributed Traffic Shaping
!
!
interface Serial1/1/0
no ip address
no fair-queue
!
interface Serial1/1/0.12 point-to-point
description 768kbps FR DLCI to RBR-3745-Left
ip address 10.1.121.1 255.255.255.252
! Binds the map-class to the FR-DLCI
!
!
! Attaches nested MQC policies to map-class
service-policy output MQC-DTS-768
! Enables FRF.12
!
show policy map

show frame-relay fragment (on slow-speed links only)
488
ATM
As with Frame Relay, ATM is an NBMA medium that permits oversubscription and speed
mismatches, and thus requires shaping to guarantee service levels. In ATM, however,
shaping is included as part of the PVC denition.
Two options exist for carrying voice trafc over slow-speed ATM PVCs: either Multilink
PPP over ATM (MLPoATM), in conjunction with MLP LFI, or ATM PVC bundling. ATM
PVC bundling is a legacy technique that has drawbacks such as inefcient bandwidth
utilization and classication limitations (IP precedence versus DSCP). But sometimes
service providers make ATM PVC bundles economically attractive to enterprise customers,
so both approaches are discussed.
Slow-Speed ( 768 kbps) ATM Links: MLPoATM

Recommendation: Use MLP LFI. Tune the ATM PVC Tx-ring to 3. cRTP can be used only
in Cisco IOS Release 12.2(2)T or later.
Serialization delays on slow-speed ATM links, as shown in Figure 13-4, necessitate a
fragmentation and interleaving mechanism. The most common ATM adaptation layers
(such as AAL5) do not have sequence numbers in the cell headers and, thus, require cells
to arrive in the correct order. This requirement makes interleaving a problem that cannot be
solved at these ATM adaptation layers and thus must be solved at a higher layer.
Figure 13-14 Slow-Speed MLPoATM Links
MLPoATM Link
768 kbps
ATM
Cloud
Branch Router
WAN Aggregator
A solution to this problem is to run MLPoATM and let MLP LFI handle any necessary
fragmentation and interleaving so that such operations are completely transparent to the
lower ATM layer. As far as the ATM layer is concerned, all cells arrive in the same order
they were sent.
MLPoATM functionality is enabled through the use of virtual-access interfaces. Virtualaccess interfaces are built on demand from virtual-template interfaces and inherit their
conguration properties from the virtual templates they are built from. Thus, the IP address,
service-policy statement, and LFI parameters all are congured on the virtual template, as
489
cRTP is supported only on ATM PVCs (through MLPoATM), as of Cisco IOS Release
12.2(2)T.
Additionally, as discussed previously in this chapter and in Chapter 5, it is recommended
that the value of the nal output buffer, the Tx-ring, be tuned on slow-speed ATM PVCs to
a value of three particles to minimize serialization delay.
Example 13-22 Slow-Speed ( 768 kbps) MLPoATM QoS Design Example
!
interface ATM4/0
bandwidth 768
no ip address
no atm ilmi-keepalive
!
pvc BRANCH#60 0/60
!
vbr-nrt 768 768
!
tx-ring-limit 3
protocol ppp Virtual-Template60
!
!
bandwidth 768
ip address 10.200.60.1 255.255.255.252
!
ppp multilink
!
!
!
NOTE
ATM PVC definition

Per-PVC Tx-ring is tuned to 3 particles
PVC is bound to the Virtual-Template
Attaches MQC policy to Virtual-Template

Enables MLP Fragmentation
Enables MLP Interleaving
When using virtual templates for low-speed ATM links, keep the following in mind:
The dynamic nature of virtual-template interfaces might make network management
unwieldy.
MLPoATM can be supported only on hardware that supports per-VC trafc shaping.
show policy map

show atm pvc
Verication Command: show atm pvc

In ATM, the length of the Tx-ring is dened in particles, not packets. The size of a particle
varies according to hardware. For example, on a Cisco 7200 PA-A3, particles are 580 bytes
490
(including a 4-byte ATM core header). This means that a 1500-byte packet would require
three particles of buffering. Furthermore, ATM denes Tx-rings on a per-PVC basis, as
shown in Examples 13-23 and 13-24.
Example 13-23 Basic ATM PVC Conguration Example
!
ip address 10.2.12.1 255.255.255.252
pvc 0/12
vbr-nrt 768 768
! ATM PVC definition
!
!
The size of a default Tx-ring can be ascertained using the show atm pvc command (an
output modier is used to focus on the relevant portion of the output), as shown in
Example 13-28.
Example 13-24 show atm pvc Verication of Tx-ring Setting
WAG-7206-Left#show atm pvc 0/12 | include TxRingLimit
VC TxRingLimit: 40 particles
The output shows that the Tx-ring is set, in this instance, to a default value of 40 particles.
The Tx-ring for the PVC can be tuned to the recommended setting of 3 using the tx-ringlimit command under the PVCs denition, as shown in Example 13-25.
Example 13-25 Tuning an ATM PVC Tx-ring
WAG-7206-Left(config)#interface atm 3/0.1
WAG-7206-Left(config-subif)#pvc 0/12
WAG-7206-Le(config-if-atm-vc)#tx-ring-limit
tx-ring-limit 3
The new setting can be veried quickly with the same show atm pvc command variation,
as shown in Example 13-25 (see Example 13-26).
Example 13-26 show atm pvc Verication of Tx-ring Setting After Tuning
WAG-7206-Left#show atm pvc 0/12 | include TxRingLimit
VC TxRingLimit: 3 particles
Slow-Speed ( 768 kbps) ATM Links: ATM PVC Bundles

Recommendation: Queuing policies for voice are not required (because voice uses a
dedicated ATM PVC). Tune the ATM PVC Tx-ring to 3.
An alternative option to provisioning QoS on slow-speed ATM PVCs is to use PVC
bundles, as illustrated in Figure 13-15. PVC bundles consist of two (or more) PVCs with
different ATM trafc contracts, grouped together in a logical association in which IPP
levels determine the PVC to which the packet will be directed. The decision to use PVC
491
bundles instead of MLPoATM for slow-speed ATM links is usually a matter of economics
(because service providers often offer attractive pricing for PVC bundles) and
conguration/management complexity comfort levels.
Figure 13-15 Slow-Speed ATM PVC Bundles
ATM PVC for Voice
ATM PVC for Data
ATM
Cloud
Branch Router
WAN Aggregator
In Example 13-27, one PVC (for voice) has a variable bit rate, non-real-time (VBR-nrt)
ATM trafc contract and an admission criterion of IPP 5, while another PVC (for data) has
an unspecied bit rate (UBR) ATM trafc contract and accepts all other precedence levels.
Again, it is also recommended that the TX-ring be tuned to 3 on such slow-speed
ATM PVCs.
Example 13-27 Slow-Speed ( 768 kbps) ATM PVC Bundles QoS Design Example
!
match ip dscp cs3
match ip dscp af31
class-map match-any Critical Data
match ip dscp cs6
match ip dscp af21
match ip dscp cs2
!
!
policy-map WAN-EDGE-DATA-PVC
! Only data queuing is required (no voice)
bandwidth percent 5
class Critical Data
class class-default
fair-queue
!
vc-class atm VOICE-PVC-256
! Voice PVC-class definition
! Voice ATM PVC definition
vbr-nrt 256 256
! Per-PVC Tx-ring is tuned to 3 particles
tx-ring-limit 3
! Only IPP5 traffic (voice) can use this PVC
precedence 5
! Traffic will not be accepted from other PVCs
no bump traffic
! Optional: Protects VC status of Voice PVC
protect vc
continues
492
Example 13-27 Slow-Speed ( 768 kbps) ATM PVC Bundles QoS Design Example (Continued)
!
vc-class atm DATA-PVC-512
! Data PVC-class definition
! Data ATM PVC definition
ubr 512
! Per-PVC Tx-ring is tuned to 3 particles
tx-ring-limit 3
! All other IPP values (data) use this PVC
precedence other
!
!
interface ATM3/0
no ip address
!
ip address 10.200.60.1 255.255.255.252
bundle BRANCH#60
pvc-bundle BRANCH60-DATA 0/60
class-vc DATA-PVC-512
! Assigns PVC to data-class
service-policy output
output WAN-EDGE-DATA-PVC
WAN-EDGE-DATA-PVC
! Attaches (data) MQC policy to PVC
service-policy
pvc-bundle BRANCH60-VOICE 0/600
class-vc VOICE-PVC-256
! Assigns PVC to voice-class
!
A major drawback to PVC bundling is that data never can get access to the voice PVC,
even if there is available bandwidth in it. This forces suboptimal consumption of WAN
bandwidth.
show policy map

show atm pvc
show atm vc
show atm bundle
Verication Command: show atm vc

The show atm vc command details the congured ATM PVCs and highlights their
encapsulation, ATM trafc contracts (or service contracts), status, and activity, as shown in
Example 13-28.
Example 13-28 show atm vc Verication of ATM PVC Denitions and Activity
WAN-AGG-7200#show atm vc
VCD /
Interface Name
VPI VCI Type
3/0.60
BRANCH60-DATA 0 60 PVC
3/0.60
BRANCH60-VOICE 0 600 PVC
WAN-AGG-7200#
Encaps
SNAP
SNAP
SC
UBR
VBR
Peak
Kbps
512
256
Avg/Min Burst
Kbps
Cells
512
1145
256
94
Sts
UP
UP
493
Verication Command: show atm bundle

The show atm bundle command provides details on the congured and current admission
criteria for individual ATM PVCs. In Example 13-29, PVC 0/600 (the voice PVC) accepts
only trafc that has been marked to IPP 5 (voice). All other IPP values (0 to 4 and 6 to 7)
are assigned to PVC 0/60 (the data PVC). This command also shows the activity for
each PVC.
Example 13-29 show atm bundle Verication of ATM PVC Bundle Denitions and Activity
WAN-AGG-7200#show atm bundle
BRANCH#60 on ATM3/0.60: UP
Config
VC Name
VPI/ VCI Prec/Exp
BRANCH60-DATA 0/60
BRANCH60-VOICE 0/600
WAN-AGG-7200#
7-6, 4-0
5
Current
Prec/Exp
Bumping PG/ Peak Avg/Min Burst

PrecExp/ PV Kbps
kbps Cells Sts
Accept
7-6, 4-0 - / Yes 512 512 1145
UP
5
- / No
PV
256 256
UP
94
Medium-Speed ( T1/E1) ATM Links

Recommendation: Use ATM inverse multiplexing over ATM (IMA) to keep future
expansion easy to manage. No LFI is required. cRTP is optional.
ATM IMA is a natural choice for medium-speed ATM links, as shown in Figure 13-16.
Although the inverse-multiplexing capabilities are not used at these speeds, IMA interfaces
make future expansion to high-speed links easy to manage (as will be demonstrated
between Example 13-30 and the high-speed ATM link in Example 13-35).
Figure 13-16 Medium-Speed ATM Links
ATM T1/E1 IMA Link
(Only One Group Member Is Active)
ATM
Cloud
Branch Router
WAN Aggregator
Example 13-30 Medium-Speed (T1/E1) ATM IMA QoS Design Example

!
interface ATM3/0
no ip address
! ATM3/0 added to ATM IMA group 0
ima-group 0
no scrambling-payload
continues
494
Example 13-30 Medium-Speed (T1/E1) ATM IMA QoS Design Example (Continued)
!
!
interface ATM3/IMA0
no ip address
!
interface ATM3/IMA0.12 point-to-point
ip address 10.200.60.1 255.255.255.252
description T1 ATM-IMA to Branch#60
pvc 0/100
! ATM PVC defined under ATM IMA sub-int
vbr-nrt 1536 1536
service-policy output WAN-EDGE ! Attaches MQC policy to PVC
!
!
show policy map

show atm pvc
show ima interface atm
High-Speed (Multiple T1/E1) ATM Links

Recommendation: Use ATM IMA and add members to the IMA group, as needed.
Previous options for accommodating multiple T1/E1 links were software-based loadsharing solutions (MLP bundling and IP CEF per-packet load sharing). As such, these
methods require additional CPU cycles to accommodate load-sharing multiple physical
links. However, with ATM IMA, inverse multiplexing over multiple T1/E1 links, illustrated
in Figure 13-17, is done in hardware on the port adaptor/network module. Therefore, ATM
IMA scales much more efciently.
Figure 13-17 High-Speed ATM Links
ATM T1/E1 IMA Link
(Multiple Active Group Members)
ATM
Cloud
Branch Router
WAN Aggregator
495
As mentioned, ATM IMA makes bandwidth expansion easy to manage. For example, all
that is required to add another T1 line to the previous example is to add an ima-group statement to the next ATM interface and increase the PVC speed, as shown in Example 13-31.
Example 13-31 High-Speed (Multiple T1/E1 and Greater) ATM IMA QoS Design Example
!
interface ATM3/0
no ip address
ima-group 0
!
interface ATM3/1
no ip address
ima-group 0
!
!
interface ATM3/IMA0
no ip address
!
ip address 10.6.12.1 255.255.255.252
pvc 0/100
! ATM PVC speed expanded
vbr-nrt 3072 3072
! Attaches MQC policy to PVC
!
!
show policy map

show atm pvc
Verication Command: show ima interface atm

The show ima interface atm command is useful for verifying that all members of an ATM
IMA group are active. See Example 13-32.
496
Example 13-32 show ima interface atm Verication of ATM IMA Group
WAG-7206-Left#show ima interface atm 3/ima0
Interface ATM3/IMA0 is up
Group index is 1
Ne state is operational, failure status is noFailure
Active links bitmap 0x3
IMA Group Current Configuration:
Tx/Rx configured links bitmap 0x3/0x3
Tx/Rx minimum required links 1/1
Maximum allowed diff delay is 25ms, Tx frame length 128
Ne Tx clock mode CTC, configured timing reference link ATM3/0
Test pattern procedure is disabled
IMA Group Current Counters (time elapsed 257 seconds):
0 Ne Failures, 0 Fe Failures, 0 Unavail Secs
IMA Group Total Counters (last 5 15 minute intervals):
0 Ne Failures, 0 Fe Failures, 0 Unavail Secs
IMA link Information:
Link
Physical Status
NearEnd Rx Status
Test Status
-------------------------------------------ATM3/0
up
active
disabled
ATM3/1
up
active
disabled
ATM3/2
administratively down unusableInhibited
disabled
ATM3/3
administratively down unusableInhibited
disabled
Very-High-Speed (DS3-OC3+) ATM Links

Recommendation: Use newer hardware platforms and keep an eye on CPU levels.
Major site-to-site interconnections drift slightly away from the traditional WAN aggregator/
remote branch router models. In site-to-site scenarios, as illustrated in Figure 13-18, the
WAN edge routers usually support only one or two links, as opposed to dozens or hundreds
of links that typical WAN aggregators support. However, in a site-to-site scenario, the
interconnecting links are running at far higher speeds than most remote branch links.
Figure 13-18 Very High-Speed (DS3-OC3+) ATM Links
WAN Edge Router
WAN Edge Router

ATM DS3/OC3 Link
ATM
Cloud
The policies and design principles do not change for site-to-site scenarios. The main
consideration is the performance of the WAN edge router. Although newer platforms handle
complex policies more efciently, it is still highly recommended that proof-of-concept
497
testing of the platforms involved be performed before implementing policies at such critical
junctions in the network. Example 13-33 illustrates a site-to-site QoS policy applied to a
very-high-speed ATM (OC3) link.
Example 13-33 Very High-Speed (DS3-OC3+) ATM Link QoS Design Example
!
interface ATM3/0
no ip address
load-interval 30
!
ip address 10.2.12.1 255.255.255.252
pvc 0/12
! ATM OC3 PVC definition
vbr-nrt 149760 149760
!
!
show policy map

show atm pvc

Many enterprises are deploying converged networks that use ATM at the central site and
Frame Relay at the remote branches. The media conversion is accomplished through
ATM-to-Frame Relay Service Interworking (SIW or FRF.8) in the carrier network.
FRF.12 cannot be used because, currently, no service provider supports FRF.12 termination
in the Frame Relay cloud. In fact, no Cisco WAN switching devices support FRF.12.
Tunneling FRF.12 through the service providers network does no good because there is no
FRF.12 standard on the ATM side. This is a problem because fragmentation is a requirement if any of the remote Frame Relay sites uses a circuit speed of 768 kbps or below.
However, MLPoATM and MLPoFR provide an end-to-end, Layer 2 fragmentation and
interleaving method for low-speed ATM to Frame Relay FRF.8 SIW links.
FRF.8 SIW is a Frame Relay Forum standard for connecting Frame Relay networks with
ATM networks. SIW provides a standards-based solution for service providers, enterprises,
and end users. In service interworking translation mode, Frame Relay PVCs are mapped to
498
ATM PVCs without the need for symmetric topologies. FRF.8 supports two modes of
operation of the interworking function (IWF) for upper-layer user protocol encapsulation:
Translation modeMaps between ATM (AAL) and Frame Relay (IETF)

encapsulation. It also supports interworking of routed or bridged protocols.
Transparent modeDoes not map encapsulations, but sends them unaltered. This
mode is used when translation is impractical because encapsulation methods do not
conform to the supported standards for service interworking.
MLP for LFI on ATM and Frame Relay SIW networks is supported for transparent-mode
VCs and translational-mode VCs that support PPP translation (FRF 8.1).
To make MLPoATM and MLPoFR SIW possible, the service providers interworking
switch must be congured in transparent mode, and the end routers must be capable of
recognizing both MLPoATM and MLPoFR headers. This is accomplished with the
protocol ppp command for ATM and the frame-relay interface-dlci dlci ppp command
for Frame Relay.
When an ATM cell is sent from the ATM side of an ATM-to-Frame Relay SIW connection,
the following must happen for interworking to be possible:
1 The sending router encapsulates a packet in the MLPoATM header by the sending
router.
2 In transparent mode, the carrier switch prepends a 2-byte Frame Relay DLCI eld to
the received packet and sends the packet to its Frame Relay interface.
3 The receiving router examines the header of the received packet. If the rst 4 bytes
after the 2-byte DLCI eld of the received packet are 0xfefe03cf, it treats it as a legal
MLPoFR packet and sends it to the MLP layer for further processing.
When a frame is sent from the Frame Relay side of an ATM-to-Frame Relay SIW
connection, the following must happen for interworking to be possible:
1 The sending router encapsulates a packet in the MLPoFR header.
2 In transparent mode, the carrier switch strips off the 2-byte Frame Relay DLCI eld
and sends the rest of the packet to its ATM interface.

3 The receiving router examines the header of the received packet. If the rst 2 bytes of
the received packet are 0x03cf, it treats it as a legal MLPoATM packet and sends it to
MLP layer for further processing.
A new ATM-to-Frame Relay SIW standard, FRF.8.1, supports MLPoATM and Frame
Relay SIW, but it could be years before all switches are updated to this new standard.
When using MLPoATM and MLPoFR, keep the following in mind:
MLPoATM can be supported only on platforms that support per-VC trafc shaping.
MLPoATM relies on per-VC queuing to control the ow of packets from the MLP
bundle to the ATM PVC.
499
MLPoATM requires the MLP bundle to classify the outgoing packets before they are
sent to the ATM VC. It also requires the per-VC queuing strategy for the ATM VC to
be FIFO because the MLP bundle handles queuing.
MLPoFR relies on the FRTS engine to control the ow of packets from the MLP
bundle to FR VC.
cRTP is supported only over ATM links (through MLPoATM), as of Cisco IOS
Release 12.2(2)T.
Slow-Speed ( 768 kbps) ATM-FR SIW Links

Recommendation: Use MLPoATM and MLPoFR. Use MLP LFI and optimize fragment
sizes to minimize cell padding. cRTP can be used only in Cisco IOS Release 12.2(2)T or
later. Tune the ATM PVC Tx-ring to 3.
As with any slow-speed WAN media, serialization delay must be addressed with a
fragmentation and interleaving mechanism. As previously mentioned, FRF.12 is not an
option for SIW links. Therefore, MLP LFI must be used. Generally, MLP LFI requires no
additional calculations to congure, but a special case exists when interworking ATM and
FR (as illustrated in Figure 13-19) because of the nature of ATMs xed cell lengths.
Figure 13-19 Slow-Speed ATM-FR SIW Links
ATM Link
768 kbps
ATM Cloud
WAN Edge Router
Frame Relay
Cloud
Frame Relay Link

768 kbps
Branch Router
When enabling MLPoATM, the fragment size should be optimized so that it ts into an
integral number of cells. Otherwise, the bandwidth required could double because of cell
padding. For example, if a fragment size of 49 bytes is congured, this fragment would
require 2 cells to transmit (because ATM cells have 48-byte payloads). This would generate
57 bytes of overhead (2 cell headers plus 47 bytes of cell padding), which is more than
double the fragment itself.
Table 13-3 provides a summary of the optimal fragment-delay parameters for MLPoATM.
500
Table 13-3
Optimal Fragment-Delay Values for MLP LFI for MLPoATM
PVC Speed
Optimal Fragment
Size
ATM Cells
(Rounded Up)
ppp multilink
fragment-delayvalue
56 kbps
84 bytes
12 ms
64 kbps
80 bytes
10 ms
128 kbps
176 bytes
11 ms
256 kbps
320 bytes
10 ms
512 kbps
640 bytes
14
10 ms
768 kbps
960 bytes
21
10 ms
A slow-speed ATM-to-Frame Relay SIW conguration is shown next, in two parts:
The central site WAN aggregator MLPoATM conguration (see Example 13-34)
The remote branch router MLPoFR conguration (see Example 13-39)
Example 13-34 MLPoATM WAN Aggregator ATM-FR SIW QoS Design Example
!
interface ATM4/0
no ip address
!
pvc BRANCH#60 0/60
!
vbr-nrt 256 256
!
tx-ring-limit 3
Tx-ring is tuned to 3 particles
protocol ppp VirtualTemplate60
!
bandwidth 256
ip address 10.200.60.1 255.255.255.252
!
ppp multilink
!
!
!
show policy map

show atm pvc
show ppp multilink
ATM PVC definition

Per-PVC
! Enables MLPoATM
Attaches MQC policy to Virtual-Template

Enables MLP fragmentation
Enables MLP interleaving
501
Example 13-35 MLPoFR Remote-Branch Router ATM-FR SIW QoS Design Example
!
interface Serial6/0
description Parent FR Link for BRANCH#60
no ip address
!
description FR Sub-Interface for BRANCH#60
bandwidth 256
! Enables MLPoFR
frame-relay interface-dlci 60 ppp Virtual-Template60
class FRTS-256kbps
!
bandwidth 256
ip address 10.200.60.2 255.255.255.252
! Attaches MQC policy to map-class
ppp multilink
! Enables MLP fragmentation
! Enables MLP interleaving
!
!
map-class frame-relay FRTS-256kbps
! CIR is set to 95% of FR DLCI rate
frame-relay cir 243200
! Bc is set to CIR/100
frame-relay bc 2432
! Be is set to 0
frame-relay be 0
! MinCIR is set to CIR
frame-relay mincir 243200
!
show policy map

show ppp multilink
ISDN
When designing VoIP over ISDN networks, special consideration needs to be given to the
following issues:
Link bandwidth varies as B channels are added or dropped.

RTP packets might arrive out of order when transmitted across multiple B channels.
CallManager has limitations with locations-based CAC.
502
Variable Bandwidth
ISDN allows B channels to be added or dropped in response to the demand for bandwidth.
The fact that the bandwidth of a link varies over time presents a special challenge to the
LLQ/CBWFQ mechanisms of Cisco IOS Software. Before Cisco IOS Release 12.2(2)T, a
policy map implementing LLQ could be assigned only a xed amount of bandwidth. On an
ISDN interface, Cisco IOS Software assumes that only 64 kbps is available, even though
the interface has the potential to provide 128 kbps, 1.544 Mbps, or 2.408 Mbps of bandwidth. By default, the maximum bandwidth assigned must be less than or equal to 75 percent of the available bandwidth. Hence, before Cisco IOS Release 12.2(2)T, only 75 percent
of 64 kbps, or 48 kbps, could be allocated to an LLQ on any ISDN interface. If more was
allocated, an error message was generated when the policy map was applied to the ISDN
interface. This severely restricted the number of VoIP calls that could be carried.
The solution to this problem was introduced in Cisco IOS Release 12.2(2)T with the
priority percent command. This command allows the reservation of a variable bandwidth
percentage to be assigned to the LLQ.
MLP Packet Reordering Considerations

MLP LFI is used for fragmentation and interleaving voice and data over ISDN links. LFI
segments large data packets into smaller fragments and transmits them in parallel across all
the B channels in the bundle. At the same time, voice packets are interleaved between the
fragments, thereby reducing their delay. The interleaved packets are not subject to MLP
encapsulation; they are encapsulated as regular PPP packets. Hence, they have no MLP
sequence numbers and cannot be reordered if they arrive out of sequence.
The packets probably will need to be reordered. The depth of the various link queues in the
bundle might differ, causing RTP packets to overtake each other as a result of the difference
in queuing delay. The various B channels also might take different paths through the ISDN
network and might end up with different transmission delays.
This reordering of packets is not generally a problem for RTP packets. The buffers on the
receiving VoIP devices reorder the packets based on the RTP sequence numbers. However,
reordering becomes a problem if cRTP is used. The cRTP algorithm assumes that RTP
packets are compressed and decompressed in the same order. If they get out of sequence,
decompression does not occur correctly.
Multiclass Multilink PPP (MCMP) offers a solution to the reordering problem. With
MCMP, the interleaved packets are given a small header with a sequence number, which
allows them to be reordered by the far end of the bundle before cRTP decompression takes
place. MCMP is supported as of Cisco IOS Release 12.2(13)T.
503
CallManager CAC Limitations

IP telephony in branch networks typically is based on the centralized call-processing model
and uses locations-based CAC to limit the number of calls across the WAN. Locationsbased CAC currently does not have any mechanism for tracking topology changes in the
network. Therefore, if the primary link to a branch goes down and ISDN backup engages,
the CallManager remains ignorant of the occurrence. For this reason, it is critical that the
ISDN backup link be capable of handling the same number of VoIP calls as the main link.
Otherwise, CAC ultimately could oversubscribe the backup link.
The actual bandwidth of the primary link and the backup link do not need to be identical.
They just need to be capable of carrying the same number of VoIP calls. For example, the
backup link might use cRTP while the primary link does not, in which case, less bandwidth
is required on the backup link to carry the same number of calls as the primary link.
Because of these limitations, it is recommended that the 33 percent LLQ recommendation
be relaxed in this kind of dial-backup scenario. The LLQ could be provisioned as high as
70 percent (leaving 5 percent for Voice control trafc over the ISDN link and 25 percent for
Best-Effort trafc).
Voice and Data on Multiple ISDN B Channels

The Voice and Data design model over ISDN, illustrated in Figure 13-20, allows a service
policy to be applied to a bundle with multiple B channels. It takes advantage of the fact that
LLQ bandwidth can be expressed as a percentage instead of an absolute number. If cRTP
is enabled, MCMP is required on the ISDN links.
Figure 13-20 Voice and Data over ISDN
ISDN Link
ISDN
Cloud
Branch Router
WAN Aggregator
Cisco IOS provides two mechanisms for controlling how channels are added in response to
demand.
The rst mechanism commonly is referred to as dial-on-demand routing (DDR). With
DDR, a load threshold must be specied (as a fraction of available bandwidth). When the
trafc load exceeds this number, an additional channel is added to the bundle. The threshold
is calculated as a running average. As a result, there is a certain delay in bringing up additional
504
B channels when the load increases. This delay does not present a problem with data, but it
is unacceptable with voice. This delay can be reduced to around 30 seconds by adding the
load-interval command to the physical ISDN interface, but even 30 seconds is too long.
The second mechanism is a more robust solution, which is simply to bring up all B channels
immediately and keep them up as long as the ISDN service is required. This is achieved by
using the ppp multilink links minimum command.
With two B channels available, the service policy can reserve (approximately) 90 kbps (70
percent of 128 kbps) for voice trafc. The total number of calls that can be transmitted
depends on the codec and sampling rates used.
Example 13-36 illustrates the conguration for enabling voice and data over multiple ISDN
B channels.
Example 13-36 Voice and Data over Multiple ISDN B Channels QoS Design Example
!
match ip dscp ef
!
match ip dscp cs3
match ip dscp af31
!
!
policy-map WAN-EDGE-ISDN
class Voice
priority percent 70
! LLQ 33% Rule is relaxed for DDR scenarios
! Enables Class-Based cRTP
bandwidth percent 5
class class-default
fair-queue
!
interface BRI0/0
encapsulation ppp
dialer pool-member 1
!
interface Dialer1
encapsulation ppp
dialer pool 1
dialer remote-name routerB-dialer1
dialer-group 1
dialer string 12345678
! Attaches MQC policy to Dialer interface
service-policy output WAN-EDGE-ISDN
ppp multilink
! Enables MLP fragmentation
! Enables MLP interleaving
! Activates both B Channels immediately
ppp multilink links minimum 2
! Enables MCMP
ppp multilink multiclass
!
505
show policy map

show ppp multilink

A ctitious company, ABC, Inc., already correctly is classifying and marking (at Layer 3
via DSCP) all 11 classes of trafc from the QoS Baseline Model within its campus. It wants
to provision QoS policies for each of these application classes over its WANs as well.
ABC, Inc., selected ATM as its primary Layer 2 medium because of its ease of expansion
for future bandwidth needs. Each remote branch is connected to the main campus by dualT1 links (at a minimum), with some larger sites using three or four T1s for connectivity.
Most remote branches are connected using Frame Relay links (through ATM-to-FR SIW);
however, some remote sites are using ATM IMA.
ABC, Inc., has selected the Cisco 7500-VIP6-80 for the WAN aggregation platform and
is provisioning sites so that the CPU utilization of any given WAN aggregator is no more
than 75 percent; it is running Cisco IOS Release 12.3 mainline code.
WAN edge QoS designs for one of the WAN aggregation routers are shown in Figure 13-22
and Example 13-37.
Figure 13-21 Case Study: QoS Baseline Policies on WAN Aggregation Router with High-Speed ATM Links
Campus
Distribution/Core
Switches
Gig0/0
WAN Aggregator
ATM3/0 (OC3)
ATM4/0 (OC3)
Gig0/1
LAN Edges
WAN Edges
ATM
Cloud
506
Example 13-37 Case Study: QoS Baseline Policies on WAN Aggregation Router with Dual-T1 ATM IMA
Links
!
ip cef distributed
! Required for C7500-VIP
!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
match ip dscp cs6
match ip dscp cs2
match ip dscp 25
class-map match-all Streaming Video
match ip dscp cs4
match ip dscp cs1
!
policy-map WAN-EDGE
class Voice
priority percent 18
priority percent 15
bandwidth percent 5
class Routing
! Bandwidth guarantee for Routing
bandwidth percent 3
class Net Mgmt
! Bandwidth guarantee for Network Management
bandwidth percent 2
! Mission-Critical data gets min 10% BW guarantee
! Applies FQ to MC-Data class (VIP Only)
fair-queue
random-detect
! Transactional Data gets min 7% BW guarantee
bandwidth percent 7
! Applies FQ to Trans-Data class (VIP Only)
fair-queue
! Enables DSCP-WRED on Transactional Data class
class Bulk Data
bandwidth percent 4
! Applies FQ to Bulk Data class (VIP Only)
fair-queue
! Enables DSCP-WRED on Bulk Data class
! Streaming-Video gets min 10% BW guarantee
class Scavenger
Summary
507
Example 13-37 Case Study: QoS Baseline Policies on WAN Aggregation Router with Dual-T1 ATM IMA
Links (Continued)
1
class class-default
fair-queue
random-detect

! Class-Default gets min 25% BW guarantee
! Applies FQ to Class-Default (VIP Only)
! Enables WRED on Class-Default
!
interface ATM3/0
no ip address
!
ip address 10.2.60.1 255.255.255.252
description Dual-T1 ATM PVC to Branch#60
pvc 0/60
! Dual-T1 ATM PVC definition
vbr-nrt 3072 3072
!
!
show policy map

show atm pvc
Summary
This chapter discussed the QoS requirements of routers performing the role of a WAN
aggregator. Specically, it addressed the need for queuing policies on the WAN edges,
combined with shaping policies when NBMA media (such as Frame Relay or ATM) are
being used, and link-specic policies, such as LFI/FRF.12 and cRTP, for slow-speed
( 768 kbps) links.
For the WAN edges, bandwidth-provisioning guidelines were considered, such as leaving
25 percent of the bandwidth for the Best-Effort class and limiting the sum of all LLQs
to 33 percent.
Three categories of WAN link speeds and their design implications were presented:
Slow-speed ( 768 kbps) links, which can support only Three- to Five-Class QoS
models and require LFI mechanisms and cRTP.
Medium-speed ( T1/E1) links, which, likewise, can support only Three- to FiveClass QoS Models but no longer require LFI mechanisms. cRTP becomes optional.
508
High-speed (multiple T1/E1 or greater) links, which can support 5- to 11-Class QoS
Models. No LFI is required on such links. cRTP likely would have a high CPU cost
(compared to realized bandwidth savings) and, as such, generally is not recommended
for such links. Additionally, some method of load sharing, bundling, or inverse
multiplexing is required to distribute the trafc across multiple physical links.
These principles then were applied to certain WAN media designsspecically, for leased
lines, Frame Relay, ATM, and ATM-to-Frame Relay SIW. The corner case of ISDN as a
backup WAN link also was considered.
Finally, a case study was presented, illustrating a complex scenario to show how these
designs can be put together.
Further Reading
Layer 3 queuing:
Class-based weighted fair queuing (Cisco IOS Release 12.0.5T):

120t/120t5/cbwfq.htm.
Low-latency queuing (Cisco IOS Release 12.0.7T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios120/120newft/120t/120t7/pqcbwfq.htm.
Distributed low-latency queuing (Cisco IOS Release 12.1.5T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios121/121newft/121t/121t5/dtllqvip.htm.
Low-latency queuing with priority percentage support (Cisco IOS Release 12.2.2T):
122t2/ftllqpct.htm.
Congestion avoidance:
MQC-based WRED (Cisco IOS Release 12.0.5T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios120/120newft/120t/120t5/cbwfq.htm.
DiffServ-compliant weighted random early detection (Cisco IOS Release 12.1.5T):

121t5/dtdswred.htm.
Distributed class-based weighted fair queuing and distributed weighted random early
detection (Cisco IOS Release 12.1.5T): http://www.cisco.com/univercd/cc/td/doc/
product/software/ios121/121newft/121t/121t5/dtcbwred.htm.
Frame Relay trafc shaping:
Class-based shaping (Cisco IOS Release 12.1.2T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios121/121newft/121t/121t2/clsbsshp.htm.
Further Reading
509
MQC-based Frame Relay trafc shaping (Cisco IOS Release 12.2.13T): http://
frqosmqc.htm.
Distributed trafc shaping (Cisco IOS Release 12.1.5T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios121/121newft/121t/121t5/dtdts.htm.
ATM PVC trafc parameters:
Conguring ATM trafc parameters: http://www.cisco.com/univercd/cc/td/doc/

product/software/ios122/122cgcr/fwan_c/wcfatm.htm#1001126.
Link fragmentation and interleaving:
MLP interleaving and queuing for Real-Time trafc (Cisco IOS Release 12.0):
http://www.cisco.com/univercd/cc/td/doc/product/software/ios120/12cgcr/dial_c/
dcppp.htm#4550.
FRF.12 (Cisco IOS Release 12.0.4T): http://www.cisco.com/univercd/cc/td/doc/

product/software/ios120/120newft/120t/120t4/120tvofr/index.htm.
Link fragmentation and interleaving for Frame Relay and ATM virtual circuits (Cisco
IOS Release 12.1.5T): http://www.cisco.com/univercd/cc/td/doc/product/software/
ios121/121newft/121t/121t5/dtlfra.htm.
Distributed link fragmentation and interleaving over leased lines (Cisco IOS Release
122newft/122t/122t8/ftdl2.htm.
Distributed link fragmentation and interleaving for Frame Relay and ATM interfaces
(Cisco IOS Release 12.2.4T): http://www.cisco.com/univercd/cc/td/doc/product/
software/ios122/122newft/122t/122t4/ftdl.htm.
Compressed Real-Time Protocol:
RTP and TCP header compression (Cisco IOS Release 12.0.7T):

120t/120t7/rtpfast.htm.
Class-based RTP and TCP header compression (Cisco IOS Release 12.2.13T):
122t13/fthdrcmp.htm.
Tx-ring:
Tx-ring tuning: http://www.cisco.com/en/US/tech/tk39/tk824/

technologies_tech_note09186a00800fbafc.shtml.
PAK_priority:
Understanding how routing updates and Layer 2 control packets are queued on an
interface with a QoS service policy: http://www.cisco.com/warp/public/105/
rtgupdates.html.
510
ISDN:
Multiclass Multilink PPP (Cisco IOS Release 12.2.13T) http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122newft/122t/122t13/ftmmlppp.htm.
Marking:
Class-based marking (Cisco IOS Release 12.1.5T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios121/121newft/121t/121t5/cbpmark2.htm.
Enhanced packet marking (Cisco IOS Release 12.2.13T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122newft/122t/122t13/ftenpkmk.htm.
This chapter discusses Branch QoS considerations and designs, including the following:
Unidirectional applications
Branch LAN edge ingress classication
Branch NBAR policies for worm identication and policing
CHAPTER
14

Chapter 13, WAN Aggregator QoS Design, discussed the QoS design recommendations
for WAN aggregators in detail. For the most part, theses designs also apply to branch routers
located at the far end of the WAN links. However, at least three unique considerations must
be made for branch router QoS design. This chapter examines in detail these considerations
and their related designs.
The rst consideration is that of unidirectional applications. Some applications, such as
Streaming-Video (whether unicast or multicast), require bandwidth allocation only on the
WAN aggregators WAN edge, not on the branch routers WAN edge. Therefore, bandwidth
allocated to unidirectional applications on the WAN aggregator WAN edge can be redistributed
among other preferential classes on the branch routers WAN edge.
Another characteristic common to branches is that trafc destined to the campus might
not be correctly marked on the branch access switches. These switches, which are usually
lower-end switches, might or might not have the capabilities to classify by Layer 3 or 4
parameters and mark DSCP values for data applications. Therefore, classication and
marking might need to be performed on the branch routers LAN edge in the ingress
direction. Furthermore, branch routers provide the capability to use NBAR to classify and
mark ows that require stateful packet inspection.
Related to classication and NBAR, another unique consideration to branch QoS design is
that branch routers are a strategic place to deploy NBAR policies for worm identication
and policing. NBAR policies can be used to identify and drop Code Red, NIMDA, SQL
Slammer, RPC DCOM/W32/MS Blaster, Sasser, and other worms.
Figure 14-1 shows the QoS policies required on a remote branch router.
514
Chapter 14: Branch Router QoS Design
Figure 14-1 Branch Router QoS Policies

Queuing/Dropping/Shaping/Link-Efficiency Policies
for Branch-to-Campus Traffic
Optional: DSCP-to-CoS Mapping Policies

for Campus-to-Branch Traffic
Branch Router
Branch Switch
VVLAN
WAN
DVLAN
WAN Edge
LAN Edge
Classification and Marking +

NBAR Worm-Dropping Policies for
Branch-to-Campus Traffic
Branch WAN Edge QoS Design

WAN edge considerations discussed in Chapter 13 apply also to the branch routers WAN
edge. This includes the following:
Link-speed categories (slow, medium, and high) and their respective design
implications
Bandwidth-provisioning guidelines (at least 25 percent for Best-Effort trafc and no

more than 33 percent for Real-Time applications)
Link-specic caveats (such as the fact that tools such as LFI and cRTP, when enabled,
must be enabled on both ends of the WAN link for them to function correctly)
Distributed platform idiosyncrasies are not as relevant for branch router designs because
these platforms rarely are deployed at remote sites (although they can be used for site-tosite links, as discussed in Chapter 13).
Unidirectional Applications
Some applications are completely symmetrical and require identical bandwidth provisioning
on both ends of the WAN link. For example, if 100 kbps of LLQ are assigned to voice in
one direction, 100 kbps of LLQ also must be provisioned for voice in the opposite direction
(assuming that the same VoIP codecs are being used in both directions, and putting aside
Branch WAN Edge QoS Design
515
for a moment multicast Music-on-Hold [MoH] provisioning). Furthermore, having symmetrical polices on both sides of the WAN links greatly simplies QoS policy deployment
and management, which is an important aspect of large-scale designs.
However, certain applications, such as Streaming-Video and multicast MoH, most often are
unidirectional. Therefore, it might be unnecessary and even inefcient to provision any
bandwidth guarantees for such trafc on the branch router for the branch-to-campus
direction of trafc ow.
Most applications lie somewhere in the middle of the scale, between the extremes of being
fully bidirectional and being completely unidirectional. Most client/server applications lie
closer to the unidirectional end of the scale because these applications usually consist of
small amounts of client-to-server trafc coupled with larger amounts of server-to-client
trafc. Such behavior can be reected in the asymmetrical bandwidth provisioning for such
types of applications.
For purely unidirectional applications, it is recommended that provisioning be removed
from the WAN edge policies on branch routers and the allocated bandwidth be redistributed
among other classes.
Branch Router WAN Edge (10-Class) QoS Baseline Model

The inclusion or exclusion of the Streaming-Video class affects only those enterprises
deploying complex QoS class models, such as the 11-Class QoS Baseline Model. When
the Streaming-Video class is removed from this model (for branch router WAN edges), the
bandwidth previously allocated to this class can be reallocated among the other preferential
data classes, as illustrated in Figure 14-2. Notice that no class is provisioned for StreamingVideo in this model, and the bandwidth assigned to it from the 11-Class WAN edge model
(from Figure 13-6 in the previous chapter) has been reassigned to the Mission-Critical Data
class and the Transactional Data class.
The conguration for such a branch router (10-class) QoS Baseline policy on a dual-T1
interface is shown in Example 14-1. Notice that there is no Streaming-Video class and that
the bandwidth for it has been distributed equally between the Mission-Critical Data class
(now at 15 percent instead of 10 percent) and the Transactional Data class (now at 12
percent instead of 7 percent).
A bandwidth statement is used on class-default, precluding the use of fair-queue on the
class for all nondistributed platforms. Also, a max-reserved-bandwidth 100 command
must be issued on the interface before the service-policy output command.
516
Figure 14-2 Branch Router (10-Class) QoS Baseline WAN Edge Model Bandwidth Allocations Example
(Dual-T1 Link Example)
Voice
18%
BestEffort
25%
InteractiveVideo
15%
Scavenger 1%
Bulk Data4%
Transactional Data
12%
Call-Signaling 5%
Routing 3%
Example 14-1 Branch Router (10-Class) QoS Baseline WAN Edge Model
!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
class-map match-all ROUTING
match ip dscp cs6
class-map match-all NET-MGMT
match ip dscp cs2
match ip dscp 25
! Recommended markings for Transactional Data
Branch Router LAN Edge QoS Design
517
Example 14-1 Branch Router (10-Class) QoS Baseline WAN Edge Model (Continued)
class-map match-all SCAVENGER
match ip dscp cs1
!
policy-map BRANCH-WAN-EDGE
class VOICE
priority percent 18
priority percent 15
bandwidth percent 5
class ROUTING
bandwidth percent 3
class NET-MGMT
bandwidth percent 2
random-detect
class BULK-DATA
bandwidth percent 4
class SCAVENGER
bandwidth percent 1
class class-default
random-detect
!
! Recommended markings for Bulk Data


! Minimal BW guarantee for Call-Signaling
! Routing class gets 3% explicit BW guarantee
! Net-Mgmt class gets 2% explicit BW guarantee
! Mission-Critical class gets min 15% BW guarantee
! Transactional-Data class gets min 12% BW guarantee
! Enables DSCP-WRED on Transactional-Data class
! Bulk Data class gets 4% BW guarantee
! Enables DSCP-WRED on Bulk-Data class
! Default class gets min 30% BW guarantee
! Enables WRED on the default class
show policy
Now that considerations and designs for the WAN edge of the branch router have been
addressed, the next section discusses the LAN edge.

The LAN edge of the branch router can have egress and ingress policies. Because you have
been dealing with egress policies since the WAN/branch discussion began, and because the
egress policies are not only optional, but also considerably simpler, they are discussed rst.
As previously mentioned, it is better to mark at Layer 3 (DSCP) instead of Layer 2 whenever
possible because Layer 2 markings are lost when the transmission medium changes. This
is the case with any Ethernet 802.1Q/p CoS values that have been set within the campus and
518
are carried over a WAN (or VPN). In some cases, network administrators prefer to have
these markings restored at the branch; DSCP-to-CoS mapping then can be performed on
the branch routers LAN edge.
In the ingress direction, the branch router might be required to perform classication and
marking of branch-to-campus trafc. This might be because the branch switch lacks the
capability to classify and mark trafc, or because the trafc consists of stateful ows that
require NBAR for classication. NBAR classication also is required at branch LAN
ingress edges to identify (and immediately drop) known worm trafc.
Each of these types of polices is discussed in more detail in the following sections.
DSCP-to-CoS Remapping
DSCP-to-CoS remapping is optional. Newer Catalyst switches perform QoS based on
internal DSCP values that are generated either by trusted DSCP markings or by trusted CoS
markings (coupled with CoS-to-DSCP mappings). In the case of legacy switches at the
branch that perform QoS strictly by preset CoS values, CoS might need to be remapped
on the branch routers LAN edge.
Enhanced packet marking (Cisco IOS Release 12.2[13]T or higher) is the optimal tool
for resetting CoS values because it uses a table. In this manner, a default DSCP-to-CoS
mapping can be used without having to congure explicitly a class-based marking policy
that matches every DiffServ class and performs a corresponding set cos function.
In both cases, keep in mind that only Ethernet trunking protocols, such as 802.1Q, carry
CoS information. Therefore, the policy works correctly only when applied to a trunked
subinterface, not to a main Ethernet interface.
Example 14-2 presents an enhanced packet marking DSCP-to-CoS conguration for a
branch LAN edge. Note that this DSCP-to-CoS remapping policy requires application to
both the voice VLAN (VVLAN) subinterface and the data VLAN (DVLAN) subinterface
on the branch routers LAN edge.
Example 14-2 Branch LAN Edge Enhanced Packet Marking for DSCP-to-CoS Remapping Example
!
ip cef
!
policy-map BRANCH-LAN-EDGE-OUT
class class-default
set cos dscp
!
!
no ip address
speed auto
duplex auto
!
! IP CEF is Required for Packet Marking
! Enables default DSCP-to-CoS Mapping
519
Example 14-2 Branch LAN Edge Enhanced Packet Marking for DSCP-to-CoS Remapping Example (Continued)
description DVLAN SUBNET 10.1.60.0
ip address 10.1.60.1 255.255.255.0
service-policy output BRANCH-LAN-EDGE-OUT
!
description VVLAN SUBNET 10.1.160.0
ip address 10.1.160.1 255.255.255.0
!
! Restores CoS for Data VLAN
! Restores CoS on Voice VLAN
show policy
Branch-to-Campus Classication and Marking

In keeping with the unofcial Differentiated Services design principle of marking trafc as
close to its source as possible, IP phones mark voice-bearer trafc (to DSCP EF) and CallSignaling trafc (currently, to DSCP AF31, but this will soon change to DSCP CS3) on the
phones themselves. Some IP/VC devices mark Interactive-Video trafc to AF41 on their
network interface cards (NICs).
However, as has already been discussed, it is not recommended that end-user PCs be trusted
to set their CoS/DSCP markings correctly because users easily can abuse this (either
unintentionally or deliberately). Therefore, application trafc that originates from untrusted
hosts should be marked on branch access switches. However, in some circumstances, this
might not be possible:
The branch access switch does not have Layer 3 or Layer 4 awareness for trafc
classication or does not support marking.
Classication needs to be performed at the application layer (through NBAR).
In such cases, DSCP classication must be performed at the ingress interface of the branch
router.
Administrators can identify and then mark application trafc based on the following
criteria:
Source or destination IP address (or subnet)Typically, destination subnets are

used when setting branch QoS policies. For example, the destination subnet for a
group of application servers can be used to identify a particular type of client-to-server
application trafc.
520
Well-known TCP/UDP portsIt is important to know whether the well-known port

is a source port or a destination port from the branch routers perspective.
NBAR protocol (for example, Citrix or KaZaa) or application subparameter (for

example, HTTP URL)NBAR Packet Description Language Modules (PDLMs)
also can be congured to identify stateful or proprietary applications, as well as
worms.
Regardless of the method used to identify the application trafc, the inbound classication
and marking policy needs to be applied only to the DVLAN subinterface. This is because
only trusted IP Telephony applications, which mark their voice trafc correctly, are admitted
onto the voice VLAN.
NOTE
If the branch access switches support access-edge policers (as described in Chapter 12,
Campus QoS Design), these likewise should be enabled on them. This adds another layer
of defense to the network, mitigating DoS/worm trafc that originates from the branch
through Scavenger-class QoS.
If the branch access switches do not support such policing, compatible policers could be
placed at the branch router access edge. This is in harmony with the principle of policing
as close to the source as possible.
Whenever possible, though, such policing should be done in Catalyst hardware rather than
Cisco IOS Software.
Although it might be tempting to use the same class-map names (such as MISSIONCRITICAL, TRANSACTIONAL-DATA, or BULK-DATA) for ingress LAN edge classication and marking policies as are in use for egress WAN edge queuing policies, this might
cause confusion in policy denition and troubleshooting. Therefore, for management and
troubleshooting simplicity, it is benecial to have similar yet descriptive names for these
new classes (for example, branch-originated trafc might have class-map names prepended
with BRANCH-). A description can also be added to the class maps with the description
command.
Source or Destination IP Address Classication

Example 14-3 shows how trafc destined to a specic subnet (10.200.200.0/24)which,
in this case, represents a server farm of proprietary Mission-Critical Data application
serverscan be identied and marked on ingress on the branch routers LAN edge.
521
Example 14-3 Branch LAN Edge Destination IP Classication and Marking Example
!
ip cef
! Required for Packet Marking
!
class-map match-all BRANCH-MISSION-CRITICAL
match access-group name MISSION-CRITICAL-SERVERS
! ACL to reference
!
policy-map BRANCH-LAN-EDGE-IN
class BRANCH-MISSION-CRITICAL
! (Interim) Recommended marking for Mission-Critical traffic
set ip dscp 25
!
...
!
no ip address
speed auto
duplex auto
!
ip address 10.1.60.1 255.255.255.0
service-policy output BRANCH-LAN-EDGE-OUT ! Restores CoS on Data VLAN
! Marks MC Data on ingress
service-policy input BRANCH-LAN-EDGE-IN
!
...
!
ip access-list extended MISSION-CRITICAL-SERVERS
permit ip any 10.200.200.0 0.0.0.255
! MC Data Server-Farm Subnet
!
show policy
show ip access-list
Verication Command: show ip access-list

When access lists are used as the ltering criteria for a class map, the show ip access-list
command is helpful in identifying whether the access list is registering matches, especially
for ACLs that have multiple lines of match criteria. This command provides granular
visibility into which lines of an access list are registering matches. In Example 14-4, 464
matches are registered against the access list.
522
Example 14-4 show ip access-list Verication of Remote Branch LAN Edge Destination IP Classication Example
BRANCH#60-C3745#show ip access-list MISSION-CRITICAL-SERVERS
Extended IP access list MISSION-CRITICAL-SERVERS
10 permit ip any 10.200.200.0 0.0.0.255 (464
464 matches)
matches
BRANCH#60-C3745#
Well-Known TCP/UDP Port Classication

Most applications can be identied by their well-known TCP/UDP ports. Some of these
ports have keywords within Cisco IOS Software to identify them when dening access lists.
NOTE
The Internet Assigned Numbers Authority (IANA) lists registered well-known and
registered application ports at http://www.iana.org/assignments/port-numbers.
Building on Example 14-4, Example 14-5 classies and marks branch-originated FTP and
e-mail trafc, both POP3 and IMAP (TCP port 143), as Bulk Data (DSCP AF11).
Example 14-5 Branch LAN Edge Well-Known Port Classication Example
!
! Required for Packet Marking
ip cef
!
class-map match-all BRANCH-BULK-DATA
match access-group name BULK-DATA-APPS
! ACL to reference
!
set ip dscp 25
class BRANCH-BULK-DATA
! Bulk data apps are marked to AF11
set ip dscp af11
!
!
no ip address
speed auto
duplex auto
!
ip address 10.1.60.1 255.255.255.0
! Restores CoS on Data VLAN
! Marks Data on ingress
!
...
523
Example 14-5 Branch LAN Edge Well-Known Port Classication Example (Continued)
!
permit ip any 10.200.200.0 0.0.0.255
!
ip access-list extended BULK-DATA-APPS
permit tcp any any eq ftp
! Identifies
permit tcp any any eq ftp-data
! Identifies
permit tcp any any eq pop3
! Identifies
! Identifies
!
FTP Control traffic

FTP Data traffic
POP3 E-mail traffic
IMAP E-mail traffic
show policy
show ip access-list
NBAR Application Classication

At the time of this writing, Cisco IOS Software included NBAR PDLMs for 98 of the most
common network applications, with the capability to dene an additional 10 applications
using custom PDLMs.
Of these protocols, 15 require stateful packet inspection for positive identication. Because
NBAR operates in the IP Cisco Express Forwarding (CEF) switching path, only the rst
packet within a ow requires stateful packet inspection, and the policy is applied to all
packets belonging to the ow. NBAR stateful packet inspection requires more CPU
processing power than simple access control lists (ACLs). However, on newer branch router
platforms, such as the Cisco 3745 or 2691, Cisco Technical Marketing testing has shown
the overhead of enabling NBAR classication at dual-T1 rates to be quite minimal (typically
2 to 5 percent, depending on the trafc mix).
Building again on the previous example, Example 14-6 uses NBAR to identify several
types of Transactional Data applications, Network-Management applications, and
Scavenger applications.
In Example 14-6, Transactional Data applications include Citrix, LDAP, Oracle SQL*NET,
and HTTP web trafc with SalesReport in the URL. In addition, a custom PDLM is
dened to identify SAP trafc (by TCP ports 3200 through 3203 and 3600), and SAP also
is included as a Transactional Data application.
Additionally, Network-Management applications, such as SNMP, Syslog, Telnet, NFS,
DNS, ICMP, and TFTP, are identied through NBAR and marked to DSCP CS2.
524
Furthermore, Scavenger applications, such as Napster, Gnutella, KaZaa (versions 1 and 2),
Morpheus, Grokster, and many other peer-to-peer le-sharing applications, are identied
by NBAR and marked to DSCP CS1.
Example 14-6 Branch LAN Edge NBAR Classication Example
!
ip nbar port-map custom-01 tcp 3200 3201 3202 3203 3600
! PDLM Mapping for SAP
!
ip cef
! IP CEF is required for both Class-Based Marking and for NBAR
!
class-map match-any BRANCH-TRANSACTIONAL-DATA!
BRANCH-TRANSACTIONAL-DATA Must use "match-any"
match protocol citrix
! Identifies Citrix traffic
match protocol ldap
! Identifies LDAP traffic
match protocol sqlnet
! Identifies Oracle SQL*NET traffic
match protocol http url "*SalesReport*"
! Identifies "SalesReport" URLs
protocol custom-01
custom-01
match protocol
! Identifies SAP traffic via Custom-01 PDLM Port-Map
class-map match-any BRANCH-NET-MGMT
match protocol snmp
! Identifies SNMP traffic
match protocol syslog
! Identifies Syslog traffic
match protocol telnet
! Identifies Telnet traffic
match protocol nfs
! Identifies NFS traffic
match protocol dns
! Identifies DNS traffic
match protocol icmp
! Identifies ICMP traffic
match protocol tftp
! Identifies TFTP traffic
class-map match-any BRANCH-SCAVENGER
match protocol napster
! Identifies Napster traffic
match protocol gnutella
! Identifies Gnutella traffic
match protocol fasttrack
! Identifies KaZaa (v1) traffic
match protocol kazaa2
!
set ip dscp 25
class BRANCH-TRANSACTIONAL-DATA
! Transactional Data apps are marked to DSCP AF21
set ip dscp af21
class BRANCH-NET-MGMT
! Network Management apps are marked to DSCP CS2
set ip dscp cs2
set ip dscp af11
class BRANCH-SCAVENGER
! Scavenger apps are marked to DSCP CS1
set ip dscp cs1
!
...
!
no ip address
speed auto
duplex auto
525
Example 14-6 Branch LAN Edge NBAR Classication Example (Continued)

!
ip address 10.1.60.1 255.255.255.0
service-policy output BRANCH-LAN-EDGE-OUT ! Restores CoS on Data VLAN
! Input Marking policy on DVLAN only
!
...
!
permit ip any 10.200.200.0 0.0.0.255
!
!
!
NOTE
The NBAR fasttrack PDLM identies KaZaa (version 1), Morpheus, Grokster, and other
applications. The NBAR gnutella PDLM identies Gnutella, BearShare, LimeWire, and
other peer-to-peer applications.
show policy
show ip access-list
show ip nbar port-map
Verication Command: show ip nbar port-map

When NBAR custom PDLMs are used to identify applications, it is useful to verify that the
ports bound to the PDLM have been entered correctly. This information is obtained with
the show ip nbar port-map command. In this (ltered) example, Example 14-7, the TCP
ports given for SAP (3200 through 3203 and 3600) were bound correctly to the Custom-01
NBAR PDLM.
526
Example 14-7 show ip nbar port-map Verication Custom-PDLM TCP/UDP Port-Mapping Example
BRANCH#60-C3745#show ip nbar port-map | include custom-01
port-map custom-01
tcp 3200 3201 3202 3203 3600
BRANCH#60-C3745#
NBAR Known-Worm Classication and Policing

Worms are nothing new. Theyve been around almost as long as the Internet itself (one of
the rst Internet worms was the Morris worm, released in November 1988). Typically,
worms are self-contained programs that attack a system and try to exploit a vulnerability in
the target. Upon successfully exploiting the vulnerability, the worm copies its program
from the attacking host to the newly exploited system to begin the cycle again.
NOTE
A virus, which is slightly different from a worm, requires a vector to carry the virus code
from one system to another. The vector can be either a word-processing document, an
e-mail, or an executable program.
The main element that distinguishes a worm from a virus is that a computer virus requires
human intervention to facilitate its spreading, whereas worms (once released) propagate
without requiring additional human intervention.
Worms are comprised of three primary components (as illustrated in Figure 14-3):
The enabling exploit codeThe enabling exploit code is used to exploit a vulnerability on a system. Exploitation of this vulnerability provides access to the system and
the capability to execute commands on the target system.
A propagation mechanismWhen access has been obtained through the enabling

exploit, the propagation mechanism is used to replicate the worm to the new target.
The method used to replicate the worm can be achieved through the use of the Trivial
File Transfer Protocol (TFTP), FTP, or another communication method. When the
worm code is brought to the new host, the cycle of infection can be started again.
A payloadSome worms also contain payloads, which might include additional

code to further exploit the host, modify data on the host, or change a web page. A
payload is not a required component, and, in many cases, the worms enabling exploit
code itself can be considered the payload.
527
Figure 14-3 Anatomy of an Internet Worm
1The Enabling
Exploit Code
2Propagation
Mechanism
3Payload
Some worms use unique TCP/UDP ports to propagate. These types of worms are fairly
simple to block using access lists (when the ports are known). Such ACLs can be congured
on the branch switch (whenever supported) or on the branch routers LAN edge.
Other worms hijack legitimate TCP/UDP ports to carry their harmful payloads. For these
latter types of worms, NBAR can be used at the branch LAN edge to perform deep-packet
analysis and drop any packets that are carrying the payloads of known worms. Some of
these known worms include Code Red, NIMDA, SQL Slammer, RPC DCOM/W32/MS
Blaster, and Sasser. The following sections discuss how NBAR can be used for each of
these types of worms.
NBAR Versus Code Red

First released in July 2001, Code Red targeted Microsoft Internet Information Server (IIS)
using a vulnerability in the IIS Indexing Service. Although the rst variant of this worm did
little damage because of a aw in the random numbergenerator code used to generate
addresses of hosts to exploit, a second variant appeared with the aw xed.
This worm, CodeRedv2, spread quickly and became the most widespread and damaging
worm to hit the Internet since the Morris worm. CodeRedv2s success as a worm relied
on the fact that the worm exploited the vulnerability in the IIS Indexing Service only as a
means of gaining access to the host. This, coupled with the wide deployment of IIS as
well as the large number of unpatched IIS web servers, contributed to the quick and wideranging spread of the worm. An estimated 360,000 hosts were infected within a period of
14 hours.
528
CodeRedv2 temporarily replaced the home page of the web servers that it struck with a new
page. Additionally, the code of the worm indicated that it was programmed to begin a
packet-ooding DoS attack against a hard-coded IP address. (At the time, this was the IP
address of the White House web server at http://www.whitehouse.gov.)
The original Code Red payload is shown in Example 14-8. The initial infection attempt
sends a large HTTP GET request to the target IIS server.
Example 14-8 Original Code Red Payload
2001-08-04 16:32:23 24.101.17.216 - 10.1.1.75 80 GET /default.ida
default.ida
NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN
NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN
NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN
NNNNNNNNNNNNNNNNNNNNNNNNNNN%u9090%u6858%ucbd3%u7801%u9090%u6858%ucbd
3%u7801%u9090%u6858%ucbd3%u7801%u9090%u9090%u8190%u00c3%u0003%u8b00%u
531b%u53ff%u0078%u0000%u00=a 403
The CodeRedv2 payload is shown in Example 14-9.

Example 14-9 CodeRedv2 Payload
2001-08-04 15:57:35 64.7.35.92 - 10.1.1.75 80 GET /default.ida
default.ida
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX%u9090%u6858%ucbd3%u7801%u
9090%u6858%ucbd3%u7801%u9090%u6858%ucbd3%u7801%u9090%u9090%u
8190%u00c3%u0003%u8b00%u531b%u53ff%u0078%u0000%u00=a 403 -
Notice that the GET request in both cases is looking for a le named default.ida. However,
this lename changes in newer variants of Code Red, such as CodeRedv3/CodeRed.C, as
Example 14-10 CodeRedv3 Payload
2001-08-06 22:24:02 24.30.203.202 - 10.1.1.9 80 GET /x.ida
x.ida
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
AAAAAAAAAAAAAAAAAAAAAAAAAAAA
AAAAAAAAAAAAAAA=X 403 HTTP/1.1 -
Although the lename has changed, the .ida sufx remains the same.
Code Red variants can include payloads that execute cmd.exe or root.exe functions to
program scripts within the IIS scripts directory, thus providing a ready-made back door to
the server for any attacker to use.
529
Therefore, to combat Code Red, NBAR policies can be congured to check the payload of
HTTP packets for these criteria (.ida, cmd.exe, and root.exe), as shown in Example 14-11.
Example 14-11 NBAR Policies to Identify Code Red
!
class-map match-any CODE-RED
match protocol http url "*.ida*"
match protocol http url "*cmd.exe*"
match protocol http url "*root.exe*"
!
! Identifies HTTP GET .ida requests

! Identifies HTTP with cmd.exe
! Identifies HTTP with root.exe
Verication command:
show policy
NBAR Versus NIMDA

Two months after Code Red struck the Internet, another large-scale worm, NIMDA, was
released. Unlike Code Red, NIMDA was a hybrid worm because it contained the
characteristics of both a worm and a virus. NIMDA spread using several vectors:
Through e-mail as an attachment (virus vector)
Through infected hosts actively scanning for back doors created by Code Red (worm
vector)
Through network shares (worm vector)

Through JavaScript by browsing compromised websites (virus vector)
Through infected hosts actively scanning for additional exploitable hosts (worm
vector)
NIMDA did not appear to exhibit intentional destructive capabilities; to date, NIMDAs
activities have been restricted to its self-propagation, which has the side effect of a DoS
(ooding) attack.
NIMDA propagates itself by copying, downloading, or executing a le called readme.eml.
Therefore, NBAR can be used to check the payload of HTTP packets to see if they are
propagating this le, as shown in Example 14-12.
Example 14-12 NBAR Policies to Identify NIMDA
!
class-map match-any NIMDA
match protocol http url "*readme.eml*"
!
Verication command:
show policy
! Identifies HTTP with "readme.eml"
530
NBAR Versus SQL Slammer

After the NIMDA infection subsided, the Internet saw the appearance of smaller infectious
worms. In January 2003, a new worm infected the Internet at such a high rate that it was
categorized as a ash worm. This worm, termed SQL Slammer, once again targeted
Microsoft Windows servers; specically, this worm targeted servers running Microsoft
Structured Query Language (SQL) Server software. The vulnerability exploited by SQL
Slammer had been published in July 2002, and a patch from Microsoft was available at that
time as well. Even though this patch was available for almost six months, SQL Slammer
spread with incredibly high efciency.
SQL slammer is a 376-byte User Datagram Protocol (UDP)based worm that infects
Microsoft SQL servers through UDP port 1434. Example 14-13 shows a signature string
from the SQL Slammer worm.
Example 14-13 SQL Slammer Worm Signature String
\x04\x01\x01\x01\x01\x01.*[.][Dd][Ll][Ll]
Because of its small size, the SQL Slammer worm is contained in a single packet. The fast
scanning rate of SQL Slammer is achieved not only because of this small size, but also
because the worm is UDP based. The worm does not have to complete a handshake
(necessary with TCP-based worms) to connect with a target system.
SQL Slammer reached its full scanning rate of 55 million scans per second within 3 minutes
of the start of the infection and infected the majority of vulnerable hosts on the Internet
within 10 minutes of the start of the infection, with an estimated 300,000 infected hosts
overall. A major consequence of such a fast scanning rate was that edge networks were
overwhelmed by the amount of trafc generated by the worm. SQL Slammers doubling
rate was approximately 8.5 seconds. In contrast, CodeRedv2s doubling rate was about
37 minutes.
SQL Slammer does not carry an additional harmful payload (beyond its enabling exploit
code), and its primary purpose is to cause DoS through exponential self propagation.
NBAR can be used to detect the SQL Slammer worm by mapping a custom PDLM to UDP
port 1434 and matching on the packet length (376-byte worm + 8 bytes of UDP header +
20 bytes of IP header = 404 bytes). This is shown in Example 14-14.
Example 14-14 NBAR Policies to Identify SQL Slammer
!
ip nbar port-map custom-02 udp 1434
!
class-map match-all SQL-SLAMMER
match protocol custom-02
match packet length min 404 max 404
!
! Maps a custom PDLM to UDP 1434
! Matches the custom Slammer PDLM

! Matches the packet length (376+28)
531
NOTE
show policy
Because NBAR custom-01 PDLM has been used in previous examples to identify SAP
trafc, another custom PDLM (custom-02) is used in this example. Subsequent examples
similarly are dened with the next-available custom PDLM.
NBAR Versus RPC DCOM/W32/MS Blaster

First released in August 2003, the Remote Procedure Call (RPC) Distributed Component
Object Model (DCOM) worm exploited a aw in a section of Microsofts RPC code dealing
with message exchange over TCP/IP, resulting in the incorrect handling of malformed
messages. This aw was a stack-based buffer overow occurring in a low-level DCOM
interface within the RPC process listening on TCP ports 135, 139, and 445.
The DCOM protocol enables Microsoft software components to communicate with one
another. This is a core function of the Windows kernel and cannot be disabled. The
vulnerability results because the Windows RPC service does not properly check message
inputs under certain circumstances. By sending a malformed RPC message, an attacker can
cause the RPC service on a device to fail in such a way that arbitrary code could be
executed. The typical exploit for this vulnerability launches a reverse-telnet back to the
attackers host to gain complete access to the target.
Successful exploitation of this vulnerability enables an attacker to run code with local system privileges. This enables an attacker to install programs; view, change, or delete data;
and create new accounts with full privileges. Because RPC is active by default on all versions of the Windows operating system, any user who can deliver a malformed TCP request
to an RPC interface of a vulnerable computer could attempt to exploit the vulnerability. It
is even possible to trigger this vulnerability through other means, such as logging into an
affected system and exploiting the vulnerable component locally.
A variant of the RPC DCOM worm is termed W32 Blaster or MS Blaster. When MS Blaster
successfully exploits a host, it attempts to upload a copy of the worm program to the newly
exploited host. MS Blaster uses TFTP to copy the worm program from the attacking host
to the target system. MS Blaster also starts up a cmd.exe process and binds it to TCP port
4444 of the newly exploited system. This provides any attacker with direct command-line
access at the local system privilege level, as discussed previously.
To access the system, the attacker needs only to telnet to TCP port 4444 on the exploited
host. If the worm is successful in copying the MS Blaster program to the target, the worm
exploit code modies the system registry to ensure that the worm is restarted if the system
532
reboots. It then launches the worm program on the newly exploited host to begin the cycle
again, starting with scanning for more exploitable hosts. MS Blaster also contained code
for a DoS attack. This particular attack was targeted at www.windowsupdate.com.
NBAR can be used to combat the RPC DCOM/W32/MS Blaster worm by identifying
communications on TCP/UDP ports 135, 139, and 445.
By default, the NBAR exchange PDLM is mapped to TCP port 135; therefore, this PDLM
can be used as part of the MS Blaster worm policy denition. Similarly, the NBAR netbios
PDLM is bound by default to TCP/UDP port 139 (in addition to TCP port 137 and UDP
ports 137 and 138), so this PDLM also can be used within the policy denition; specically,
the netbios PDLM can have its port mapping expanded to include TCP port 445 and UDP
ports 135, 139, and 445, as shown in Example 14-15.
NOTE
Alternatively, a custom PDLM can be dened for these ports (TCP/UDP 135, 139, and
445), but before this could be done, you would have to map the exchange and netbios
PDLMs ports away from their defaults, to avoid conicting PDLM port mappings.
Example 14-15 NBAR Policies to Identify RPC DCOM/W32/MS Blaster

!
ip nbar port-map netbios tcp 137 139 445
! Matches TCP 137/139/445
ip nbar port-map netbios udp 135 137 138 139 445
! Matches UDP 135/137-139/445
!
class-map match-any MS-BLASTER
match protocol exchange
! Matches TCP port 135
match protocol netbios
! Matches MS Blaster NetBIOS PDLM
!
show policy
NBAR Versus Sasser

The next major worm after MS Blaster was the Sasser worm (and variants Sasser.A/B/C/D),
which was released in late April 2004. Sasser exploits a aw in the Windows Local Security
Authority Service Server (LSASS) that can cause systems to crash and continually reboot,
or allow a remote attacker to execute arbitrary code with local system privileges.
Sasser is very efcient in scanning: It can scan 1024 separate IP addresses simultaneously
(on TCP port 445). When scanning reveals a vulnerable system, the worm exploits the
533
LSASS vulnerability and creates a remote shell (RSH) session on TCP port 9996 back to
the infecting system. Then Sasser starts an FTP server on TCP port 5554 to retrieve a copy
of the worm.
Sasser can be identied through a custom NBAR PDLM listening for communication on
TCP ports 445, 5554, and 9996, as shown in Example 14-16.
NOTE
If TCP port 445 already has been bound to the netbios NBAR PDLM (as recommended
previously in the MS Blaster worm denition), it is not necessary to include this port in the
Sasser custom PDLM port mapping (because it will cause a conict).
Example 14-16 NBAR Policies to Identify Sasser

!
ip nbar port-map custom-03 tcp 445 5554 9996
!
class-map match-all SASSER
!
! Matches on TCP 445/5554/9996
! Matches Sasser custom PDLM
show policy
NBAR Versus Future Worms

There is every reason to believe that new worms will be released in the future. These worms
will be not only more complex, but also more efcient in their propagation, and thus more
damaging in their scope.
A new NBAR feature (introduced in Cisco IOS Release 12.3[4]T) enables network
administrators to extend the capability of NBAR to classify (and monitor) additional static
port applications or to allow NBAR to classify unsupported static port trafc. Specically,
it enables administrators to dene the strings that they want to search for in the application
payload (for any application, not just HTTP URLs) to identify a given application.
This functionality can be used to identify proprietary applications that otherwise could
not be matched. However, it also can be very useful in plugging holes that future worms
might open.
534
For example, consider the example of a ctitious worm called Moonbeam. Moonbeam
scans and propagates itself on randomly generated TCP ports within the range of 21000
through 21999. Furthermore, the worm carries the word Moonbeam within the payload,
beginning with the ninth ASCII character of the string. Moonbeams (ctitious) payload is
Example 14-17 Moonbeam Worm Payload
\x04\x01Moonbeam\x01\x01\x01\x01.*[.][Dd][Ll][Ll]u9090%u6858%ucbd3%
Moonbeam
u7801%u9090%u6858%ucbd3%u7801%u9090%u6858%ucbd3%u7801%u9090%u9090%
u8190%u00c3%u0003%u8b00%u531b%u53ff%u0078%u0000%u00=a 403 ...
Moonbeam could be identied by a custom NBAR PDLM that examines TCP packets
within the range of 21000 through 21999, ignores the rst eight ASCII values, and checks
for the string Moonbeam (case sensitive), as shown in Example 14-18.
Example 14-18 NBAR Policies to Identify Moonbeam
!
ip nbar custom MOONBEAM 8 ascii Moonbeam tcp range 21000 21999 ! "Moonbeam" PDLM
!
class-map match-all MOONBEAM-WORM
match protocol MOONBEAM
! Matches the "Moonbeam" custom PDLM
!
show policy
Policing Known Worms

These are just a few examples of known worms that can be identied using NBAR. After
trafc generated by known worms has been positively identied, it should not be re-marked
or limited; rather, it should be dropped immediately. This can be done on ingress on branch
LAN edges, as shown in Example 14-19, which combines the policies for Code Red,
NIMDA, SQL Slammer, RPC DCOM/W32/MS Blaster, and Sasser. Additionally, the
ctitious worm Moonbeam has been included in the policy.
NOTE
A recursive classication is required in this policy for SQL Slammer. This is because
SQL Slammer requires a match-all criteria for its initial classication, but for policymanagement purposes, it is desired that this initial classication of SQL Slammer be
lumped under a single policy (with a match-any criteria) to identify and drop all known
worms.
535
Example 14-19 NBAR Branch LAN Edge Ingress Policy for Known Worms
!
! SQL Slammer custom PDLM
! Sasser custom PDLM
ip nbar port-map custom-03 tcp 5554 9996
! MS Blaster TCP 137/139/445
ip nbar port-map netbios udp 135 137 138 139 445 ! MS Blaster UDP 135/137-139/445
!
! Matches the SQL Slammer PDLM
!
class-map match-any WORMS
! CodeRed
! CodeRed
! CodeRed
! NIMDA
! SQL Slammer class-map
match class-map SQL-SLAMMER
! MS Blaster (TCP 135)
! MS Blaster NetBIOS PDLM
! "Moonbeam" PDLM
!
policy-map WORM-DROP
class WORMS
! Drops all known worms
drop
!
...
!
no ip address
speed auto
duplex auto
!
ip address 10.1.60.1 255.255.255.0
service-policy input WORM-DROP
! Drops known worms (DVLAN only)
!

Continuing the example from the previous chapters, the ctitious company ABC, Inc.,
already is correctly classifying and marking all 11 classes of trafc from the QoS Baseline
Model within its campus. Such marking is being performed at Layer 3 via DSCP. Furthermore, dual-rate policing is used within the campus so that certain AF classes of trafc can
be marked down to a second level of drop preference (for example, AF21 can be marked
536
down to AF22). Additionally, the WAN aggregator is provisioning LLQ and CBWFQ policies (with WRED on major data classes) for all 11 classes of trafc over the preferred
Layer 2 WAN medium: ATM.
Every quarter, ABC, Inc., multicasts company meetings to all its branches and provides
each branch with on-demand (unicast) e-learning content. However, because all the IP/TV
servers belonging to ABC, Inc., are located at the central campus, they see no need to
provision a class for Streaming-Video in both directions on the WAN links. Therefore, the
company has chosen to eliminate this class from the branch router WAN edge congurations
and redistribute the bandwidth between the Mission-Critical Data class and the Transactional
Data class.
Currently, the branch access switches do not have the capability to classify and mark trafc,
although they plan to eventually deploy switches in their branches with such capabilities
(including dual-rate policing and marking). In the meantime, ABC, Inc., will mark all
branch-to-campus trafc on the branch routers LAN edge on ingress. When the budget
enables them to upgrade their branch switches, marking policies dependent on Layer 3 or
Layer 4 criteria will be pushed out to the branch access switch (based on designs covered
in Chapter 12). However, policies will remain on the branch routers LAN edge to classify
and mark applications that require stateful packet inspection (using NBAR).
ABC, Inc., also has been hit with a number of worms over the past few years and wants to
do everything possible to limit and mitigate such attacks. The company has provisioned
Scavenger-class QoS within the campus and WAN edges. Furthermore, ABC, Inc., has
chosen to deploy NBAR policies to identify and drop known worms that might originate
within their branches. These policies are included in the branch routers ingress QoS policies.
QoS designs for the WAN and LAN edges of one of ABC, Inc.s, branch routers is
illustrated in Figure 14-4 and detailed in Example 14-20.
Figure 14-4 Case Study: Branch Router with 10-Class QoS Baseline WAN Edge Policies and DSCP-to-CoS
Remapping and NBAR Classication Plus Worm-Dropping LAN Edge Policies
10-Class LLQ/CBFQ/WRED Policies for
Policy-Map: BRANCH-WAN-EDGE
DSCP-to-CoS Mapping Policies for Campus-to-Branch Traffic

(VVLAN and DVLAN)
Policy-Map: BRANCH-LAN-EDGE-OUT
Branch Router
ATM
Cloud
Branch Switch
WLAN
Fa0/0.160
ATM 3/0
Fa0/0.60
DVLAN
ATM 3/1
WAN Edge
LAN Edge
Classification (ACL+NBAR) + Marking + Worm-Dropping Policies

(DVLAN Only)
Policy-Map: BRANCH-LAN-EDGE-IN
537
Example 14-20 Case Study: Branch Router with 10-Class QoS Baseline WAN Edge Policies, DSCP-to-CoS
Remapping, and NBAR Classication Plus Worm-Dropping LAN Edge Policies

!
! SAP custom PDLM
ip nbar port-map custom-01 tcp 3200 3201 3202 3203 3600
! SQL Slammer PDLM
! Sasser PDLM
ip nbar port-map custom-03 tcp 5554 9996
! MS Blaster NetBIOS
ip nbar port-map netbios udp 135 137 138 139 445
! MS Blaster NetBIOS
!
!
! MC Data ACL to reference
!
class-map match-any BRANCH-TRANSACTIONAL-DATA
match protocol citrix
! Identifies Citrix traffic
match protocol ldap
! Identifies LDAP traffic
match protocol sqlnet
! Identifies Oracle SQL*NET traffic
match protocol http url "*SalesReport*"
! Identifies "SalesReport" URLs
! Identifies SAP traffic via PDLM
!
class-map match-any BRANCH-NET-MGMT
! Identifies SNMP traffic
match protocol snmp
match protocol syslog
! Identifies Syslog traffic
match protocol telnet
! Identifies Telnet traffic
! Identifies NFS traffic
match protocol nfs
! Identifies DNS traffic
match protocol dns
match protocol icmp
! Identifies ICMP traffic
match protocol tftp
! Identifies TFTP traffic
!
! Reference ACL for Bulk Data apps
!
class-map match-any BRANCH-SCAVENGER
! Identifies Napster traffic
match protocol napster
match protocol gnutella
! Identifies Gnutella traffic
match protocol fasttrack
match protocol kazaa2
!
! Matches the custom Slammer PDLM
!
class-map match-any WORMS
! CodeRed
! CodeRed
! CodeRed
! NIMDA
match class-map SQL-SLAMMER
! SQL Slammer class-map
! MS Blaster (TCP 135)
! MS Blaster NetBIOS PDLM
continues
538
Remapping, and NBAR Classication Plus Worm-Dropping LAN Edge Policies (Continued)

! "Moonbeam" PDLM
!
!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
match ip dscp cs6
class-map match-all NET-MGMT
! Net-Mgmt apps are marked via NBAR ingress policy
match ip dscp cs2
match ip dscp 25
! MC Data apps are marked via ACL ingress policy
! Transactional apps are marked via NBAR policy
! Bulk Data apps are marked via ACL ingress policy
match ip dscp cs1
! Scavenger apps are marked via NBAR ingress policy
!
!
policy-map BRANCH-LAN-EDGE-OUT
! Enhanced Packet Marking DSCP-to-CoS policy
class class-default
! Default DSCP-to-CoS marking
set cos dscp
!
!
! BRANCH LAN Edge Ingress Marking Policy
! Marks Mission-Critical apps to DSCP 25
set ip dscp 25
class BRANCH-TRANSACTIONAL-DATA
! Marks Transactional Data apps to DSCP AF21
set ip dscp af21
class BRANCH-NET-MGMT
set ip dscp cs2
! Marks Net Mgmt apps to DSCP CS2
! Marks Bulk Data apps to DSCP AF11
set ip dscp af11
class BRANCH-SCAVENGER
set ip dscp cs1
! Marks Scavenger apps to DSCP CS1
class WORMS
drop
! Drops all known worms
class class-default
set ip dscp default
! Marks everything else to DSCP 0
!
!
policy-map BRANCH-WAN-EDGE
class VOICE
priority percent 18
539
priority percent 15
bandwidth percent 5
class ROUTING
bandwidth percent 3
class NET-MGMT
bandwidth percent 2
random-detect
class BULK-DATA
bandwidth percent 4
class SCAVENGER
bandwidth percent 1
class class-default
random-detect

! Minimal BW guarantee for Call-Signaling
! Routing class gets 3% explicit BW guarantee
! Net-Mgmt class gets 2% explicit BW guarantee
! Mission-Critical class gets min 15% BW guarantee
! Transactional-Data class gets min 12% BW guarantee
! Enables DSCP-WRED on Transactional-Data class
! Bulk Data class gets 4% BW guarantee
! Enables DSCP-WRED on Bulk-Data class
! Default class gets min 30% BW guarantee
! Enables WRED on the default class
!
...
!
no ip address
speed auto
duplex auto
!
ip address 10.1.60.1 255.255.255.0
! Restores CoS on Data VLAN
! Marks data and drops worms
!
ip address 10.1.160.1 255.255.255.0
!
...
!
interface ATM3/0
no ip address
ima-group 0
continues
540
!
interface ATM3/1
no ip address
ima-group 0
!
...
!
interface ATM3/IMA0
no ip address
!
ip address 10.6.12.1 255.255.255.252
pvc 0/100
vbr-nrt 3072 3072
! ATM PVC speed set for Dual-T1
service-policy output BRANCH-WAN-EDGE
! Attaches MQC policy to the ATM PVC
!
!
...
!
permit ip any 10.200.200.0 0.0.0.255
! MC Data Server-Farm Subnet
!
! Identifies FTP Control traffic
! Identifies FTP Default traffic
! Identifies POP3 E-mail traffic
! Identifies IMAP E-mail traffic
!
show policy map

show atm pvc
Further Reading
541
Summary
Although the QoS design recommendations for branch routers are very similar to and are
related to the QoS recommendations for WAN aggregators, this chapter examined three
unique considerations of branch routers.
The rst consideration is whether applications provisioned over the WAN are bidirectional
or unidirectional. Some unidirectional applications, such as Streaming-Video, provisioned
on the WAN aggregator WAN edge do not need to be provisioned correspondingly on the
branch routers WAN edge. Bandwidth from unidirectional application classes can be redistributed among other preferential application classes on the branch routers WAN edge.
The second consideration unique to branch routers is that ingress marking might need to be
performed on branch-to-campus trafc. This might be because the remote branch access
switch does not have the capability to classify and mark trafc, or it might be because some
applications require stateful packet inspection (NBAR) to identify them correctly. In either
case, ingress marking policies would be required on the branch routers LAN edge, on the
data VLANs subinterface (the voice VLAN trafc markings are trusted). Branch-to-campus
trafc can be identied by Layer 3 parameters (such as destination subnet), Layer 4 parameters (such as well-known TCP/UDP ports), or NBAR PDLMs. An example of each type
of classication is provided. Additionally, optional DSCP-to-CoS mapping policies (for
restoring 802.1p CoS markings that were lost when campus-originated trafc traversed the
WAN media) can be set on the branch routers LAN edge.
A third unique consideration in branch QoS design is that branch router ingress LAN edges
are a strategic place to deploy NBAR policies for worm identication and policing. NBAR
policies can be used to identify and drop Code Red, NIMDA, SQL Slammer, RPC DCOM/
W32/MS Blaster, Sasser, and other worms. This chapter discussed an extension of the
NBAR feature that enables administrators to program the strings that they want NBAR to
search packet payloads for; this feature enables NBAR policies to be used to identify new
worms that undoubtedly will be released in the future.
Finally, a case study was presented illustrating how these three unique considerations could
be combined in a complex branch router design.
Further Reading
Cisco IOS documentation:
Class-based marking (Cisco IOS Release 12.1.5T): http://www.cisco.com/univercd/

cc/td/doc/product/software/ios121/121newft/121t/121t5/cbpmark2.htm.
Enhanced packet marking (Cisco IOS Release 12.2.13T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122newft/122t/122t13/ftenpkmk.htm.
NBAR overview (Cisco IOS Release 12.3): http://www.cisco.com/univercd/cc/td/

doc/product/software/ios122/122cgcr/fqos_c/fqcprt1/qcfclass.htm#1003102.
542
Network-Based Application Recognition (Cisco IOS Release 12.1.5T):

fqos_c/fqcprt1/qcfnbar.htm.
Network-Based Application Recognition (Cisco IOS Release 12.2.8T):

122t/122t8/dtnbarad.htm.
Network-Based Application Recognition (Cisco IOS Release 12.3[4]T):

122t/122t8/dtnbarad.htm.
Cisco SAFE whitepapers:
SAFE worm mitigation: http://www.cisco.com/en/US/netsol/ns340/ns394/ns171/

ns128/networking_solutions_white_paper09186a00801e120c.shtml.
Using Network-Based Application Recognition and ACLs for blocking the Code
Red worm: http://www.cisco.com/en/US/products/hw/routers/ps359/
products_tech_note09186a00800fc176.shtml.
How to protect your network against the NIMDA virus: http://www.cisco.com/en/US/

products/sw/iosswrel/ps1835/products_tech_note09186a0080110d17.shtml.
SAFE SQL Slammer worm attack mitigation: http://www.cisco.com/en/US/netsol/

ns340/ns394/ns171/ns128/networking_solutions_white_paper09186a00801cd7f5.shtml.
SAFE RPC DCOM/W32/Blaster attack mitigation: http://www.cisco.com/en/

US/netsol/ns340/ns394/ns171/ns128/networking_solutions_white_
paper09186a00801b2391.shtml.
Combating the Internet worm Sasser: http://www.cisco.com/application/pdf/en/us/

guest/netsol/ns441/c664/cdccont_0900aecd800f613b.pdf.
PA R T
VPN QoS Design

Part V of this book provides an in-depth discussion of QoS considerations and designs for both MPLS
and IPSec VPNs. MPLS VPN QoS design is examined from both the enterprise subscribers perspective,
covering important considerations and designs for mapping into service-provider models, and also from
the service providers perspective, including MPLS DiffServ Tunneling modes and MPLS Trafc
Engineering (in brief). IPSec VPN designs are considered for both site-to-site scenarios and
telecommuter scenarios.
Chapter 15
MPLS VPN QoS Design
Chapter 16
MPLS VPN QoS design typically is viewed from two distinct perspectives:
The enterprise customer subscribing to the MPLS VPN service

The service provider provisioning edge and core QoS within the MPLS VPN service
To achieve end-to-end service levels, enterprise and service-provider QoS policies must be
consistent and complimentary. Therefore, QoS considerations and design recommendations
for both the enterprise and service provider are presented in this chapter. The following
topics are discussed:
Enterprise-to-service provider mapping models

Service provider-to-enterprise models
MPLS DiffServ tunneling modes
DiffServ in the backbone
MPLS trafc engineering
CHAPTER
15
MPLS VPN QoS Design

MPLS is a combination of routing and switching technologies that can provide scalable
VPNs with end-to-end quality of service.
Many customers are turning to service providers that offer MPLS VPN services as private
WAN alternatives. One of the main reasons for this is the any-to-any connectivity capabilities of MPLS VPNs. However, this full-mesh nature in itself poses signicant QoS implications to enterprise customers and service providers alikenamely, that they both need to
comanage QoS in a cooperative and complementary fashion to achieve end-to-end service
levels.
This chapter examines in detail QoS considerations that enterprise customers need to bear
in mind when subscribing to MPLS VPNs, including how best to map into various serviceprovider MPLS VPN QoS models.
Service provider-edge QoS considerations are reviewed in depth, including egress queuing
models and MPLS DiffServ tunneling modes (Uniform, Short Pipe, and Pipe). Furthermore, service-provider core QoS considerations are reviewed, including aggregate bandwidth provisioning and DiffServ in the backbone. MPLS trafc engineering as it relates to
QoS is covered, along with two detailed examples: MPLS per-VPN trafc engineering and
MPLS DiffServ trafc engineering.
This chapter concludes with a case study that shows how these designs can be combined in
a complex MPLS VPN end-to-end scenario.
NOTE
This chapter addresses QoS design for MPLS VPNs, not the theory and operation of MPLS
VPNs themselves. It is assumed that the reader is familiar with basic MPLS VPN architectures and technologies. For a detailed discussion of MPLS VPNs, refer to the Cisco Press
books MPLS and VPN Architectures, Volumes I and II, by Ivan Pepelnjak and Jim Guichard; Trafc Engineering with MPLS, by Eric Osborne and Ajay Simha; and Advanced
MPLS Design and Implementation, by Vivek Alwayn.
548
Chapter 15: MPLS VPN QoS Design

MPLS VPN architectures are comprised of customer edge (CE) routers, provider-edge (PE)
routers, and provider (P) routers. MPLS VPNs provide fully meshed Layer 3 virtual WAN
services to all interconnected CE routers, as outlined by RFC 2547. This fully meshed
characteristic of MPLS VPNs presents a signicant design implication to traditional Layer 2
WAN QoS design.
Because of cost, scalability, and manageability constraints, traditional private WAN designs
rarely use full-mesh models. Instead, most Layer 2 WAN designs revolve around a hub-andspoke model, implementing either a centralized hub design or the more efcient regional
hub design. Under such hub-and-spoke designs, QoS primarily is administered at the hub
router by the enterprise. As long as the service provider meets the contracted service levels,
the packets received at remote branches will reect the scheduling policies of the hub router
(sometimes referred to as a WAN aggregator). The WAN aggregator controls not only campus-to-branch trafc, but also branch-to-branch trafc (which is homed through the hub).
Under traditional hub-and-spoke models, QoS principally is administered by the enterprise
customer, as shown in Figure 15-1.
Figure 15-1 QoS Administration in Traditional Hub-and-Spoke Layer 2 WAN Design
Enterprise WAN Aggregators (Hubs)
Principally Control all QoS
Branch-to-Branch Traffic
WAN
Access
Switch
Distribution
Switch
Core WAN Aggregator

Switch
Router
Branch
Router
Access
Switch
However, with the advent of MPLS VPN service offerings that inherently offer full-mesh
connectivity, the QoS administration paradigm shifts. Under a full-mesh design, the hub
router still administers QoS for all campus-to-branch trafc, but it no longer fully controls
the QoS for branch-to-branch trafc. Although it might appear that the only required
workaround for this new scenario is to ensure that QoS is provisioned on all branch routers,
this is insufcient because it addresses only part of the issue.
For example, consider the case of provisioning any-to-any videoconferencing. As with a
traditional Layer 2 WAN design, a scheduling policy to prioritize IP/VC on the WAN aggregator is required. Then the enterprise must properly provision similar priority scheduling
549
for IP/VC on the branch routers also. In this manner, any videoconferencing calls from the
campus to the branch (and also from branch to branch) are protected against trafc of lesser
importance owing between the same sites. The complexity of the fully meshed model arises when considering that contending trafc might not always come for the same sites, but
could come from any site. Furthermore, the enterprise no longer fully controls QoS for
branch-to-branch trafc because this trafc no longer is homed through a hub. Continuing
the example, if a videoconferencing call is set up between two branches and a user from
one of the branches also initiates a large FTP download from the central site, the potential
for oversubscription of the PE-to-CE link from the fully meshed MPLS VPN cloud into one
of the branches becomes very real, likely causing drops from the IP/VC call.
The only way to guarantee service levels in such a scenario is for the service provider to
provision QoS scheduling that is compatible with the enterprises policies on all PE links
to remote branches. This is what creates the paradigm shift in QoS administration for fully
meshed topologies. Namely, enterprises and service providers must cooperate to jointly
administer QoS over MPLS VPNs, as shown in Figure 15-2.
Figure 15-2 QoS Administration in Fully Meshed MPLS VPN Design
Service providers (PE) principally
control Branch-to-Branch QoS.
Enterprise (CE) Edges principally
control Campus-to-Branch QoS.
Branch-to-Branch Traffic
MPLS VPN
Access
Switch
Distribution
Switch
Core
Switch
CE
Router
PE
Routers
CE
Routers
Queuing policies are mandatory on CE and PE routers because of the full-mesh implications of MPLS VPNs. PE routers also typically have policing (and markdown) policies on
ingress to enforce SLAs.
QoS policies on P routers are optional. Such policies are optional because some service
providers overprovision their MPLS core networks and, as such, do not require any
additional QoS policies within their backbones; on the other hand, other providers might
implement simplied DiffServ policies within their cores or might even deploy MPLS
trafc engineering (MPLS TE) to handle congestion scenarios within their backbones. Figure
15-3 summarizes the points where QoS policies can be provisioned within MPLS VPN
architectures.
550
Figure 15-3 Where QoS Is Required in MPLS VPN Architectures

CE-to-PE Queuing/Shaping/Remarking/LFI
Optional: Core DiffServ or MPLS TE Policies
PE Ingress Policing and Remarking
P Routers
CE Router
PE Router
PE Router
CE Router
MPLS VPN
PE-to-CE Queuing/Shaping/LFI
Required
Optional
This design chapter rst examines CE and PE QoS designs; then it overviews MPLS VPN
core QoS options and designs.
Customer Edge QoS Design Considerations

In addition to the full-mesh implication of MPLS VPNs, these considerations should be
kept in mind when considering MPLS VPN CE QoS design:
Layer 2 access (link-specic) QoS design

Service-provider service-level agreements (SLA)
Enterprise-to-service provider mapping models
The following sections examine these considerations in more detail.
Layer 2 Access (Link-Specic) QoS Design

Although MPLS VPNs are essentially Layer 3 WANs, a Layer 2 access medium to connect
to the MPLS VPN service provider is an obvious requirement. Most providers support
Frame Relay and ATM as access media because this makes migration from Layer 2 WANs
to Layer 3 MPLS VPNs easier and cheaper to manage; customers are not forced to convert
551
hardware on hundreds (or, in some cases, thousands) of remote branch routers to connect
to MPLS VPN providers.
It is important to recognize that Layer 2 QoS link-specic issues and designs remain the
same with regular Layer 2 WAN edges or with Layer 3 MPLS VPN CE/PE edges. For
example, shaping and LFI recommendations for slow-speed FR links are identical whether
the link is used for a Layer 2 WAN or for a Layer 3 MPLS VPN access link. Again, this
makes migration easier to manage because link-specic QoS designs do not need to be
changed (although the service policy itself might require minor modication, which is
discussed in more detail shortly).
In addition to FR and ATM for access, some service providers support Ethernet/Fast
Ethernet as access media but usually guarantee a CIR of only subline rate. In such cases,
hierarchical shaping and queuing policies on the CE edges are recommended, as illustrated
later in this chapter.
Service-Provider Service-Level Agreements

End-to-end QoS is like a chain that is only as strong as the weakest link. Therefore, its
essential for enterprises (with converged networks) subscribing to MPLS VPN services to
choose service providers that can provide the required SLAs for their converged networks.
For example, these are the end-to-end SLA requirements of voice and interactive video:
No more than 150 ms of one-way latency from mouth to ear (per ITU G.114 standard)
No more than 30 ms of jitter
No more than 1 percent loss
As a subset of the trip, the service providers component of the SLA must be considerably
tighter. These SLAs are dened for Cisco-Powered Networks (CPN)IP Multiservice
Service Providers:
No more than 60 ms of one-way latency from edge to edge

No more than 20 ms of jitter
No more than 0.5 percent loss
Figure 15-4 illustrates the interrelationship of these SLAs.

CPN-IP Multiservice Service Providers that meet these SLAs can be found at http://
www.cisco.com/pcgi-bin/cpn/cpn_pub_bassrch.pl; choose the IP VPN Multiservice option.
To achieve such end-to-end SLAs, enterprise customers (managing CEs) and service
providers (managing PEs and core Ps) must cooperate and be consistent in classifying,
provisioning, and integrating their respective QoS designs. To this end, various mapping
models have been developed to integrate enterprise requirements into service-provider
solutions.
552
Figure 15-4 CPN-IP Multiservice Service-Provider SLAs
Maximum One-Way End-to-End Service-Levels

Latency 150 ms / Jitter 30 ms / Loss 1%
Enterprise
Campus
Service Provider
Enterprise
Branch
Maximum One-Way
Edge-to-Edge
SP Service-Levels
Latency 60 ms
Jitter 20 ms
Loss 0.5%
Enterprise-to-Service Provider Mapping Models

Although Cisco is adopting its new QoS Baseline initiative and designing tools such as
Cisco AutoQoS Enterprise to facilitate and simplify the deployment of advanced QoS
trafc models within the enterprise, to date, very few enterprises have deployed more than
a handful of trafc classes. Therefore, most service providers offer only a limited number
of classes within their MPLS VPN clouds. At times, this might require enterprises to
collapse the number of classes that they have provisioned to integrate into their service
providers QoS models. The following caveats should be remembered when deciding how
best to collapse and integrate enterprise classes into various service-provider QoS models.
553
Voice and Video

Service providers typically offer only one Real-Time class or Priority class of service. If an
enterprise wants to deploy both Voice and Interactive-Video (each of which is recommended
to be provisioned with strict priority treatment) over their MPLS VPN, they might be faced
with a dilemma. Which one should be assigned to the Real-Time class? Are there any implications about assigning both to the Real-Time class?
Keep in mind that voice and video should never both be assigned low-latency queuing on
link speeds where serialization is a factor ( 768 kbps). Packets offered to the LLQ typically
are not fragmented; thus, large IP/VC packets can cause excessive delays for VoIP packets
on slow-speed links.
An alternative is to assign IP/VC to a nonpriority class, which entails not only the obvious
caveat of lower service levels, but also possible trafc-mixing concerns, as discussed
shortly.
Call-Signaling
VoIP requires provisioning not only of RTP bearer trafc, but also of Call-Signaling trafc,
which is very lightweight and requires only a moderate amount of guaranteed bandwidth.
Because the service levels applied to Call-Signaling trafc directly affect delay to the dial
tone, it is important from the end users expectations that Call-Signaling be protected.
Service providers might not always offer a suitable class just for call signaling trafc itself,
leading to the question of which other trafc classes Call-Signaling should be mixed with.
On links where serialization is not an issue (> 768 kbps), Call-Signaling could be provisioned into the Real-Time class, along with voice.
However, this is not recommended on slow-speed links where serialization is a factor. On
such slow-speed links, Call-Signaling is best assigned to one of the preferential data classes
for which the service provider provides a bandwidth guarantee.
It is important to realize that a guarantee applied to a service-provider class as a whole does
not itself guarantee adequate bandwidth for an individual enterprise applications within
the class.
Mixing TCP with UDP

It is a general best practice to not mix TCP-based trafc with UDP-based trafc (especially
Streaming-Video) within a single service-provider class because of the behaviors of these
protocols during periods of congestion. Specically, TCP transmitters throttle back ows
when drops are detected. Although some UDP applications have application-level windowing, ow control, and retransmission capabilities, most UDP transmitters are completely
oblivious to drops and, thus, never lower transmission rates because of dropping.
554
When TCP ows are combined with UDP ows within a single service-provider class and
the class experiences congestion, TCP ows continually lower their transmission rates,
potentially giving up their bandwidth to UDP ows that are oblivious to drops. This effect
is called TCP starvation/UDP dominance.
TCP starvation/UDP dominance likely occurs if (TCP-based) Mission-Critical Data is
assigned to the same service-provider class as (UDP-based) Streaming-Video and the class
experiences sustained congestion. Even if WRED is enabled on the service-provider class,
the same behavior would be observed because WRED (for the most part) manages
congestion only on TCP-based ows.
Granted, it is not always possible to separate TCP-based ows from UDP-based ows, but
it is benecial to be aware of this behavior when making such application-mixing decisions
within a single service-provider class.
Marking and Re-Marking

Most service providers use Layer 3 marking attributes (IPP or DSCP) of packets offered to
them to determine which service provider class of service the packet should be assigned to.
Therefore, enterprises must mark or re-mark their trafc consistent with their service
providers admission criteria to gain the appropriate level of service. Additionally, service
providers might re-mark at Layer 3 out-of-contract trafc within their cloud, which might
affect enterprises that require consistent end-to-end Layer 3 markings.
A general DiffServ principle is to mark or trust trafc as close to the source as administratively and technically possible. However, certain trafc types might need to be re-marked
before handoff to the service provider to gain admission to the correct class. If such
re-marking is required, it is recommended that the re-marking be performed at the CEs
egress edge, not within the campus. This is because service-provider service offerings likely will evolve or expand over time, and adjusting to such changes will be easier to manage
if re-marking is performed only at the CE egress edge.
Additionally, in some cases, multiple types of trafc are required to be marked to the same
DiffServ code point value to gain admission to the appropriate queue. For example, on
high-speed links, it might be desired to send Voice, Interactive-Video, and Call-Signaling
to the service providers Real-Time class. If this service-provider class admits only DSCP
EF and CS5, two of these three applications would be required to share a common code
point. The class-based marking conguration in Example 15-1 shows how this can be done
(in this example, both Interactive-Video and Call-Signaling are re-marked to share DSCP
CS5).
Example 15-1 CE (Egress) Enterprise-to-Service Provider Re-Marking Example
!
class-map match-any VOICE
match ip dscp ef
555
Example 15-1 CE (Egress) Enterprise-to-Service Provider Re-Marking Example (Continued)

match ip dscp af41
match ip dscp af31
match ip dscp cs3
!
policy-map CE-EGRESS-EDGE
class VOICE
priority percent 18
priority percent 15
set ip dscp cs5
! Interactive-Video is remarked to CS5
priority percent 2
! Call-Signaling gets LLQ for this scenario
set ip dscp cs5
! Call-Signaling is also remarked to CS5
!
!
interface Serial1/0
service-policy output CE-EGRESS-EDGE
!
show policy
Service providers might re-mark trafc at Layer 3 to indicate whether certain ows are out
of contract. Although this is consistent with DiffServ standards, such as RFC 2597, it might
present minor difculties to enterprises that require consistent end-to-end Layer 3 marking
(typically, for management or accounting purposes). In such cases, the enterprise can
choose to apply re-marking policies as trafc is received back from the service providers
MPLS VPN (on the ingress direction of the enterprises CE).
Class-based marking can be used again because it supports not only access lists for
classication, but also Network-Based Application Recognition (NBAR).
Continuing and expanding on the previous example, the enterprise wants to restore the original markings that it set for Interactive-Video and Call-Signaling. Additionally, it wants to
restore original markings for Oracle trafc (which it originally marked DSCP 25 and is
using TCP port 9000 with) and DLSw+ trafc (originally marked AF21). Both of these data
applications were handed off to the service provider marked as AF21, but they might have
been marked down to AF22 within the service-provider cloud. Example 15-2 shows a
conguration that enables such re-marking from the MPLS VPN. The match-all criteria
of the class maps performs a logical AND operation against the potential markings and
re-markings, and the access list (or NBAR-supported protocol) that sifts the applications
apart. The policy is applied on the same CE link, but in the ingress direction.
556
Example 15-2 CE (Ingress) Service Provider-to-Enterprise Re-Marking Example

!
class-map match-all REMARKED-INTERACTIVE-VIDEO
match ip dscp cs5
! Interactive-Video must be CS5 AND UDP
!
class-map match-all REMARKED-CALL-SIGNALING
match ip dscp cs5
! Call-Signaling must be CS5 AND TCP
!
class-map match-all REMARKED-ORACLE
! Oracle may have been remarked to AF22
! Oracle uses TCP port 9000
!
class-map match-all REMARKED-DLSW+
! DLSw+ may have been remarked to AF22
! DLSw+ is identified by NBAR
match protocol dlsw
!
policy-map CE-INGRESS-EDGE
class REMARKED-INTERACTIVE-VIDEO
! Restores Interactive-Video marking to AF41
set ip dscp af41
class REMARKED-CALL-SIGNALING
! Restores Call-Signaling marking to AF31
set ip dscp af31
class REMARKED-ORACLE
! Restores Oracle marking to DSCP 25
set ip dscp 25
class REMARKED-DLSW+
! Restores DLSw+ marking to AF21
set ip dscp af21
!
!
interface serial 1/0
service-policy output CE-EGRESS-EDGE
service-policy input CE-INGRESS-EDGE
! Marking restoration on ingress
!
!
access-list 101 permit udp any any
! Identifies UDP traffic
access-list 102 permit tcp any any
! Identifies TCP traffic
access-list 103 permit tcp any eq 9000 any
! Identifies Oracle on TCP 9000
!
show policy
Three-Class Provider-Edge Model: CE Design

In this model, the service provider offers three classes of service: Real-Time (strict priority,
available in 5-percent increments), Critical Data (guaranteed bandwidth), and Best-Effort.
The admission criterion for the Real-Time class is either DSCP EF or CS5; the admission
557
criterion for Critical Data is DSCP CS6, AF31, or CS3. All other code points are re-marked
to 0. Additionally, out-of-contract AF31 trafc can be marked down within the service
providers MPLS VPN cloud to AF32.
Under such a model, there is no recommended provision for protecting Streaming-Video
(following the Dont mix TCP with UDP guideline), nor is there a service-provider class
suitable for bulk data, which consists of large, nonbursty TCP sessions that could drown out
smaller data transactions. Figure 15-5 shows a re-marking diagram for a three-class serviceprovider model.
Figure 15-5 Three-Class Provider-Edge Model Re-Marking Diagram
Enterprise
Application
DSCP
Routing
CS6
Voice
EF
Interactive-Video
Streaming-Video
AF41
PE Classes
EF
CS5
CS4
DSCP 25
AF31
Call-Signaling
AF31/CS3
CS5
Transactional Data
AF21
CS3
Network-Management
CS2
CS3
Bulk Data
AF11
Scavenger
CS1
Best-Effort
CS5
Real-Time
35%
CS6
AF31
CS3
Critical
Data
40%
X
Best-Effort
25%
Example 15-3 shows an example CE conguration for an advanced enterprise model

mapping (over a dual-T1 link) into a three-class service-provider model.
Example 15-3 CE Conguration for Three-Class Provider-Edge Model
!
class-map
match ip
class-map
match ip
match-all ROUTING
dscp cs6
match-all VOICE
dscp ef
continues
558
Example 15-3 CE Conguration for Three-Class Provider-Edge Model (Continued)

match ip dscp af41
match ip dscp 25
match ip dscp af31
match ip dscp cs3
match ip dscp af21
class-map match-all NETWORK-MANAGEMENT
match ip dscp cs2
match ip dscp cs1
!
!
policy-map CE-THREE-CLASS-SP-MODEL
class ROUTING
bandwidth percent 3 ! Routing is assigned (by default) to Critical SP class
class VOICE
priority percent 18 ! Voice is admitted to Realtime SP class
priority percent 15
! Interactive-Video is assigned to the Realtime SP class
set ip dscp cs5
priority percent 2
! Call-Signaling is assigned to the Realtime SP class
set ip dscp cs5
random-detect
! MC Data is assigned to the Critical SP class
set ip dscp af31
random-detect
! Transactional Data is assigned to Critical SP class
set ip dscp cs3
class NETWORK-MANAGEMENT
bandwidth percent 2
! Net Mgmt is assigned to Critical SP class
set ip dscp cs3
class SCAVENGER
bandwidth percent 1
class class-default
random-detect
!
559
show policy
The max-reserved-bandwidth command might be required on the interface to which the

previously discussed policy is applied.
Four-Class Provider-Edge Model: CE Design

Building on the previous model, a fourth class is added that can be used for either Bulk Data
or Streaming-Video. The admission criterion for this new class is either DSCP AF21 or
CS2. The re-marking diagram shown in Figure 15-6 illustrates how this new class can be
used for Streaming-Video and (primarily UDP-based) Network-Management trafc.
Figure 15-6 Four-Class Provider-Edge Model Re-Marking Diagram
Enterprise
Application
DSCP
Routing
CS6
Voice
EF
PE Classes
EF
Interactive-Video
AF41
CS5
Streaming-Video
CS4
AF21
DSCP 25
AF31
Call-Signaling
AF31/CS3
CS5
Transactional Data
AF21
CS5
CS6
AF31
CS3
CS3
Network-Management
CS2
Bulk Data
AF11
Scavenger
CS1
Best-Effort
AF21
CS2
Real-Time
35%
Critical
Data
25%
Video
15%
X
Best-Effort
25%
Example 15-4 shows an example CE conguration for an advanced enterprise model

mapping (over a dual-T1 link) into a four-class service-provider model.
560
Example 15-4 CE Conguration for Four-Class Provider-Edge Model

!
match ip dscp cs6
match ip dscp ef
match ip dscp af41
class-map match-all STREAMING-VIDEO
match ip dscp cs4
match ip dscp 25
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp cs2
match ip dscp cs1
!
!
policy-map CE-FOUR-CLASS-SP-MODEL
class ROUTING
class VOICE
priority percent 15
set ip dscp cs5
class STREAMING-VIDEO
! Streaming-Video is assigned to the Video SP class
set ip dscp af21
priority percent 2
set ip dscp cs5
random-detect
set ip dscp af31
random-detect
set ip dscp cs3
bandwidth percent 2 ! Net Mgmt (mainly UDP) is admitted to Video SP class
class SCAVENGER
bandwidth percent 1
class class-default
random-detect
!
561
show policy

previously discussed policy is applied.
Five-Class Provider-Edge Model: CE Design

Building again on the previous model, a fth class is added that also can be used for either
Bulk Data or Streaming-Video (whichever wasnt used under the four-class model). The
admission criterion for this new class is either DSCP AF11 or CS1, which necessitates the
previously unrequired re-marking of the Scavenger class to DSCP 0 (so that it will not be
admitted into the Bulk Data class, but will fall into the Best-Effort class). Figure 15-7
illustrates the re-marking required when using this new class for Bulk Data.
Figure 15-7 Five-Class Provider-Edge Model Re-Marking Diagram
Enterprise
Application
DSCP
Routing
CS6
Voice
EF
PE Classes
EF
Interactive-Video
AF41
CS5
Streaming-Video
CS4
AF21
DSCP 25
AF31
Call-Signaling
AF31/CS3
CS5
Transactional Data
AF21
CS3
Network-Management
CS2
Bulk Data
AF11
Scavenger
CS1
Best-Effort
CS5
Real-Time
35%
CS6
AF31
CS3
Critical
Data
20%
AF21
CS2
Video
15%
AF11/CS1 Bulk Data 5%

0
Best-Effort
25%
Example 15-5 shows an example CE conguration for a QoS Baseline enterprise model
mapping (over a dual-T1 link) into a ve-class service-provider model.
562
Example 15-5 CE Conguration for Five-Class Provider-Edge Model

!
match ip dscp cs6
match ip dscp ef
match ip dscp af41
match ip dscp cs4
match ip dscp 25
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp af11
match ip dscp cs2
match ip dscp cs1
!
!
policy-map CE-FIVE-CLASS-SP-MODEL
class ROUTING
class VOICE
priority percent 15
set ip dscp cs5
set ip dscp af21
priority percent 2 ! Call-Signaling gets LLQ for this scenario
set ip dscp cs5
random-detect
set ip dscp af31
bandwidth percent 5
random-detect
set ip dscp cs3
class BULK-DATA
bandwidth percent 5 ! Bulk Data is assigned to Bulk SP class
random-detect
class SCAVENGER
Provider-Edge QoS Considerations
563
Example 15-5 CE Conguration for Five-Class Provider-Edge Model (Continued)

bandwidth percent 1
set ip dscp 0
class class-default
random-detect
! Scavenger is re-marked to 0
show policy

preceding policy is applied.

PE designs are relevant for service providers (and for enterprises that are self-managing
their own MPLS VPNs). Two unique considerations for PE QoS design are discussed next:
Service provider-to-enterprise models

MPLS DiffServ tunneling modes
These considerations are examined in more detail in the following sections.
Service Provider-to-Enterprise Models

The PE edges facing customer CEs are complementary to the enterprise-to-service provider
mapping models discussed previously. The PE designs for each class model (three, four,
and ve) are detailed in the following sections.
Three-Class Provider-Edge Model: PE Design

As outlined previously (and illustrated in Figure 15-5), in this model, the service provider
offers three classes of service: Real-Time (strict priority, available in 5-percent increments),
Critical Data (guaranteed bandwidth), and Best-Effort. The admission criterion for the
Real-Time class is either DSCP EF or CS5; the admission criterion for Critical Data is
DSCP CS6 (for customer routing trafc), AF31, or CS3. All other code points are remarked to 0 by an ingress policer (not shown in this conguration example, but detailed
later under the MPLS DiffServ tunneling examples). Additionally, service-provider
policers can re-mark out-of-contract AF31 trafc down to AF32, which results in a higher
drop preference because DSCP-based WRED is enabled on this class. As in previous
examples, Example 15-6 is based on an access link of more than 3 Mbps.
564
Example 15-6 PE Conguration for Three-Class Provider-Edge Model

!
class-map match-any REALTIME
match ip dscp ef
match ip dscp cs5
class-map match-any CRITICAL-DATA
match ip dscp cs6
match ip dscp af31
match ip dscp cs3
!
policy-map PE-THREE-CLASS-SP-MODEL
class REALTIME
! Realtime class gets 35% LLQ
priority percent 35
class CRITICAL-DATA
! Critical-Data SP class gets 40% CBWFQ
! DSCP-based WRED enabled on class
class class-default
! Best Effort SP class gets FQ
fair-queue
! WRED enabled on Best Effort SP class
random-detect
!
show policy
Four-Class Provider-Edge Model: PE Design

Building on the previous model (and as illustrated in Figure 15-6), a fourth class is added
to this SP model, which can be used for either Bulk Data or Streaming-Video. The
admission criterion for this new class is either DSCP AF21 or CS2. Out-of-contract AF21
trafc offered to this class can be marked down to AF22. In this particular example, the
class is being called Video, but it is important to keep in mind that the customer can offer
any trafc desired to this class, provided that it is marked appropriately. For this reason
(although it normally is not required on UDP-based ows such as Streaming-Video),
DSCP-based WRED is enabled on this class to aggressively drop out-of-contract trafc
as needed. As in previous examples, Example 15-7 is based on an access link of more than
3 Mbps.
Example 15-7 PE Conguration for Four-Class Provider-Edge Model
!
class-map
match ip
match ip
class-map
match ip
match ip
match ip
match-any REALTIME
dscp ef
dscp cs5
match-any CRITICAL-DATA
dscp cs6
dscp af31
dscp cs3
565
Example 15-7 PE Conguration for Four-Class Provider-Edge Model (Continued)

class-map match-any VIDEO
match ip dscp af21
match ip dscp cs2
!
policy-map PE-FOUR-CLASS-SP-MODEL
class REALTIME
! Realtime SP class gets 35% LLQ
priority percent 35
class CRITICAL-DATA
class VIDEO
! Video SP class gets 15% CBWFQ
! DSCP-based WRED enabled on Video SP class
class class-default
! Best Effort SP class gets FQ
fair-queue
random-detect
!
show policy
Five-Class Provider-Edge Model: PE Design

Building again on the previous model (and as illustrated in Figure 15-7), a fth class is
added that can be used for either Bulk Data or Video (whichever wasnt used under the fourclass model). In this example, the new class is used for Bulk Data. The admission criterion
for this new class is either DSCP AF11 or CS1. Out-of-contract AF11 trafc offered to this
class can be re-marked to AF12 and can be discarded earlier by the DSCP-based WRED
algorithm operating on the output queue for this class.
To prevent long TCP sessions of the Bulk Data SP class from dominating bandwidth intended for the Best-Effort class, a bandwidth guarantee is offered to the Best-Effort class. This
guarantee might require the use of the max-reserved-bandwidth override under the applied
interface conguration. As in the previous examples, an access link of more than 3 Mbps is
assumed in Example 15-8.
Example 15-8 PE Conguration for Five-Class Provider-Edge Model
!
class-map
match ip
match ip
class-map
match ip
match ip
match-any REALTIME
dscp ef
dscp cs5
match-any CRITICAL-DATA
dscp cs6
dscp af31
continues
566
Example 15-8 PE Conguration for Five-Class Provider-Edge Model

match ip dscp cs3
match ip dscp af21
match ip dscp cs2
class-map match-any BULK-DATA
match ip dscp af11
match ip dscp cs1
!
policy-map PE-FIVE-CLASS-SP-MODEL
class REALTIME
priority percent 35
class CRITICAL-DATA
class VIDEO
class BULK-DATA
! Bulk Data SP class gets 15% CBWFQ
bandwidth percent 5
! DSCP-based WRED enabled on Bulk Data SP class
class class-default
! Best Effort SP class gets 25% CBWFQ
random-detect
!
show policy
MPLS DiffServ Tunneling Modes

As described in previous examples, some service providers re-mark packets at Layer 3 to
indicate whether trafc is in contract or out-of-contract. Although this conforms to DiffServ
standards, such as RFC 2597, this is not always desirable from an enterprise customers
standpoint.
Because MPLS labels include 3 bits that commonly are used for QoS marking, it is possible
to tunnel DiffServthat is, preserve Layer 3 DiffServ markings through a service providers MPLS VPN cloud while still performing re-marking (via MPLS EXP bits) within
the cloud to indicate in- or out-of-contract trafc.
RFC 3270 denes three distinct modes of MPLS DiffServ tunneling; each is discussed in
detail in the following sections:
Uniform Mode
Short Pipe Mode
Pipe Mode
567
Uniform Mode
Uniform Mode generally is utilized when the customer and service provider share the same
DiffServ domain, as in the case of an enterprise deploying its own MPLS VPN core.
In Uniform Mode, which is the default mode, the rst 3 bits of the IP ToS eld (IP
Precedence bits) automatically are mapped to the MPLS EXP bits on the ingress PE as
labels are pushed onto the packets.
If policers or any other mechanisms re-mark the MPLS EXP values within the MPLS core,
these marking changes are propagated to lower-level labels and eventually are propagated
to the IP ToS eld (MPLS EXP bits are mapped to IP Precedence values on the egress PE).
Figure 15-8 shows the behavior of Uniform Mode MPLS DiffServ tunneling.
Figure 15-8 MPLS DiffServ Uniform Tunneling Mode Operation
Assume a policer re-marks out-of-contract
traffics topmost label to MPLS EXP 0 here.
MPLS VPN
CE Router
PE Router
PE Router
CE Router
P Routers
IPP3/DSCP AF31
Packet initially
marked to IPP3/
DSCO AF31.
MPLS EXP 3
MPLS EXP 0
MPLS EXP 3
MPLS EXP 0
IPP3/DSCP AF31
MPLS EXP 3
IPP3/DSCP AF31
By default, IPP values
will be copied to
Topmost label is
MPLS EXP labels.
marked down by a
policer.
IPP3/DSCP AF31
Topmost label is
popped and EXP value
is copied to underlying
label.
IPP0/DSCP 0
MPLS EXP value
is copied to
IP ToS byte.
Direction of Packet Flow
The mapping of IP Precedence to MPLS EXP is performed by default on PEs for customerto-provider trafc.
568
However, for provider-to-customer egress trafc (from the MPLS VPN cloud), additional
conguration is required on the PE to achieve mapping of MPLS EXP to IP Precedence.
This is because the nal label is popped (and discarded) when it is received from the MPLS
VPN cloud and, therefore, cannot be used as a match criterion for policies applied to the
egress interface of the nal PE router (facing the destination CE). The solution is to copy
the nal MPLS EXP bit values to a temporary placeholder on PE ingress from the MPLS
core (before the label is discarded) and then use these temporary placeholder values for
setting the IP Precedence bits on egress to the customer CE.
Cisco IOS provides two such temporary placeholders, the QoS Group and the Discard
Class. For Uniform Mode scenarios, it is recommended to copy the MPLS EXP values to
QoS Group values on ingress from the MPLS VPN cloud. (The Discard Class is recommended for use in Pipe Mode scenarios only.) Then QoS Group values can be copied to IP
Precedence values (on egress to the customer CE). Figure 15-9 illustrates the policies required
for a single direction for Uniform Mode MPLS DiffServ tunneling. (This policy also would
be required on the complementary interfaces for the reverse trafc direction.)
Figure 15-9 MPLS DiffServ Uniform Tunneling Mode Policies
MPLS VPN
CE Router
(Ingress)
PE Router
The MPLS EXP value from the final

label is temporarily copied to a QoS
Group on ingress from the MPLS VPN
cloud and then the label is popped
(and discarded).
(Egress)
PE Router
CE Router
P Routers
IP Precedence values
are copied to MPLS EXP
labels by default.
The QoS Group value is copied

to the IP ToS byte on egress,
setting the IP Precedence to
match the last MPLS EXP value.
Example 15-9 shows the conguration for Uniform Mode operation on a PE.
569
Example 15-9 PE Conguration for MPLS DiffServ Uniform Mode Tunneling

!
policy-map MPLSEXP-TO-QOSGROUP
class class-default
set qos-group mpls experimental topmost ! Copies MPLS EXP to QoS Group
!
policy-map QOSGROUP-TO-IPP
class class-default
set precedence qos-group
! Copies QoS Group to IPP
!
!
interface ATM2/0
no ip address
!
description ATM-OC3 TO MPLS VPN CORE
! Link to/from MPLS VPN Core
ip address 20.2.34.4 255.255.255.0
pvc 0/304
vbr-nrt 149760 149760
service-policy input MPLSEXP-TO-QOSGROUP ! MPLS EXP to QoS Group on ingress
!
tag-switching ip
!
!
description FE TO CUSTOMER RED CE
! Link to/from CE
ip vrf forwarding RED
ip address 10.1.45.4 255.255.255.0
service-policy output QOSGROUP-TO-IPP
! QoS Group to IPP on egress to CE
!
show policy
Of course, additional QoS policies (to these Uniform Mode tunneling policies), such as
queuing or WRED, can be applied on the PE-to-CE egress link (as detailed earlier in the
previous section).
Short Pipe Mode

Short Pipe Mode is utilized when the customer and service provider are in different
DiffServ domains. (The service providers DiffServ domain begins at the ingress PEs
ingress interface and terminates on the egress PEs ingress interface.)
570
This mode is useful when the service provider wants to enforce its own DiffServ policy and
the customer requests that its DiffServ information be preserved through the MPLS VPN
cloud. Short Pipe Tunneling Mode provides DiffServ transparency through the service
provider network (as does Pipe Mode).
The outmost label is utilized as the single most meaningful information source as it relates
to the service providers QoS PHB. On MPLS label imposition, the IP classication is not
copied into the outermost labels EXP. Instead, the value for the MPLS EXP is set explicitly
on the ingress PEs ingress interface, according to the service providers administrative
policies.
In the case of any re-marking occurrence within the service providers MPLS VPN cloud,
changes are limited to MPLS EXP re-marking only and are not propagated down to the
underlying IP packets ToS byte. Figure 15-10 shows the operation of Short Pipe Mode
MPLS DiffServ tunneling.
Figure 15-10 MPLS DiffServ Short Pipe Mode Tunneling Operation
Shaded area represents Provider DiffServ domain.
Unshaded areas represent
Customer DiffServ domain.

PE Edge (to CE) policies

are based on
customer markings.
MPLS VPN
CE Router
PE Router
PE Router
CE Router
P Routers
IPP3/DSCP AF31
Packet initially
marked to IPP3/
DSCO AF31.
MPLS EXP 4
MPLS EXP 0
MPLS EXP 4
MPLS EXP 0
IPP3/DSCP AF31
MPLS EXP 4
MPLS EXP values

are set independently
from IPP/DSCP values.
IPP3/DSCP AF31
Topmost label is
marked down by a
policer.
IPP3/DSCP AF31
IPP3/DSCP AF31
Original customermarked IP ToS

Topmost label is
values are
popped and EXP value preserved.
label.
571
MPLS EXP values can be marked in any way that the provider wants to provide local
signicance. Figure 15-11 shows an example use of MPLS EXP markings to indicate in- or
out-of-contract trafc for a ve-class service-provider model.
Figure 15-11 Five-Class Service Provider Model Short Pipe Mode Re-Marking Diagram
PE Classes
Real-Time In Contract
EF
CS5
CE-to-PE
packets marked
with appropriate
DSCP values to
gain admission
into desired
SP Class.
MPLS EXP values are set on ingress PEs

ingress interface to reflect local SP policies.
Real-Time
MPLS EXP 5
Real-Time Out-of-Contract
(Dropped)
CS6
AF31
CS3
Critical
Data
AF21
CS2
Video
AF11
CS1
Bulk Data
Critical Data In Contract

Critical Data Out-of-Contract
MPLS EXP 7
Video In Contract
MPLS EXP 2
Video Out-of-Contract
(Dropped)
Bulk Data In Contract
Bulk Data Out-of-Contract
Best-Effort In Contract
MPLS EXP 3
Best-Effort
Best-Effort Out-of-Contract
MPLS EXP 1
MPLS EXP 6
MPLS EXP 0
MPLS EXP 4
Figure 15-12 shows the ingress PE ingress interface re-marking policies for Short Pipe
Mode, based on the re-marking diagram provided in Figure 15-11. No mapping from
MPLS EXP to QoS Group is needed on the egress PEs ingress interface (as was required
for Uniform Mode) because the MPLS EXP value loses relevance beyond this interface.
Any egress policies on the egress PEs egress interface (facing the customers destination
CE), are based on IP Precedence or DSCP values (which have remained untouched). This
is the main difference between Short Pipe Mode and Pipe Mode.
Figure 15-12 shows the interfaces in which explicit policy conguration is required for
Short Pipe Mode MPLS DiffServ tunneling.
572
Figure 15-12 MPLS DiffServ Short Pipe Mode Tunneling Policies
MPLS VPN
CE Router
(Ingress)
PE Router
(Egress)
PE Router
CE Router
P Routers
MPLS EXP values are

explicitly set by
service provider.
Egress policies based on

customers DiffServ markings.
Example 15-10 shows the conguration for Short Pipe Mode operation on a PE. Trafc
received from CEs is marked explicitly (through MPLS EXP values) to reect the service
providers policies. In this example, the customer is given a 3-Mbps CIR through an FE
access link. The provider is using a ve-class model with 35 percent for Real-Time trafc,
20 percent for Critical Data trafc, 15 percent for Video trafc, 5 percent for Bulk Data
trafc, and 25 percent for Best-Effort trafc. On PE-to-CE links (in the egress direction),
queuing and dropping policies based on customer IP DiffServ markings also are
recommended (as was discussed previously).
Example 15-10 PE Conguration for MPLS DiffServ Short Pipe Mode Tunneling
!
match ip dscp ef
match ip dscp cs5
match ip dscp cs6
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp cs2
match ip dscp af11
match ip dscp cs1
!
!
policy-map PE-FIVE-CLASS-SHORT-PIPE-MARKING
573
Example 15-10 PE Conguration for MPLS DiffServ Short Pipe Mode Tunneling (Continued)
class REALTIME
police cir 1050000
exceed-action drop
class CRITICAL-DATA
police cir 600000
exceed-action set-mpls-exp-topmost-transmit 7
class VIDEO
police cir 450000
exceed-action drop
class BULK-DATA
police cir 150000
class class-default
police cir 750000
! Conforming RT set to 5
! Excess Realtime is dropped
! Critical Data set to 3

! Excess Critical set 7
! Conforming Video set to 2

! Excess Video dropped
! Conforming Bulk set to 1

! Excess Bulk set to 6
! Conforming BE set to 0
! Excess BE set to 4
!
! Link to/from CE
ip address 10.1.12.2 255.255.255.0
service-policy input PE-FIVE-CLASS-SHORT-PIPE-MARKING
!
show policy
Pipe Mode
The main difference between Short Pipe Mode and Pipe Mode MPLS DiffServ tunneling
is that the PE egress policies (toward the customer CEs) are provisioned according to the
service providers explicit markings and re-markings, not the enterprise customers IP
DiffServ markings (although these are preserved). As with Short Pipe Mode, any changes
to label markings that occur within the service providers cloud do not get propagated to the
IP ToS byte when the packet leaves the MPLS network.
Because egress PE-to-CE QoS policies in Pipe Mode are dependent on the last MPLS EXP
value, this value must be preserved before the nal label is popped. A temporary placeholder (as used in Uniform Mode operation) is again required. On the nal PE router in
574
a given path, the MPLS EXP value is copied to the QoS Group value. Optionally, a Discard
Class value also might set drop preference at the same time. Thereafter, egress queuing or
dropping policies are performed based on these QoS Group/Discard Class values. Figure 15-13
illustrates the Pipe Mode MPLS DiffServ tunneling operation.
Figure 15-13 MPLS DiffServ Pipe Mode Tunneling Operation
Assume a policer re-marksout-of-contract


are based on
provider markings.
MPLS VPN
CE Router
PE Router
PE Router
CE Router
P Routers
IPP3/DSCP AF31
Packet initially
marked to IPP3/
DSCO AF31.
MPLS EXP 4
MPLS EXP 0
MPLS EXP 4
MPLS EXP 0
IPP3/DSCP AF31
MPLS EXP 4
IPP3/DSCP AF31
MPLS EXP values
are set independently Topmost label is
from IPP/DSCP values marked down by a
policer.
IPP3/DSCP AF31
Topmost label is
label.
IPP3/DSCP AF31
values are
preserved.
QoS Groups and Discard Classes can be combined to provide virtual DiffServ PHB
classication. For example, RFC 2597 assured-forwarding PHBs can be mimicked using
QoS Group values 1 through 4 (to represent the AF class) coupled with Discard Class
values 1 through 3 (to represent the drop preference). In general, QoS Group and Discard
Class values are arbitrary and have only local signicance. However, an exception is found
when WRED is congured to selectively drop based on Discard Class values, in which case
the lower Discard Class values are dropped rst (by default). If no Discard Class value is
assigned explicitly, the value defaults to 0.
Figure 15-14 shows the points where policies are required for Pipe Mode MPLS DiffServ
tunneling.
575
Figure 15-14 MPLS DiffServ Pipe Mode Tunneling Policies

group (and Discard class) on ingress
from the MPLS VPN cloud and then
the label is popped (and discarded).
MPLS VPN
CE Router
(Ingress)
PE Router
(Egress)
PE Router
CE Router
P Routers
MPLS EXP values are

explicitly set by
service provider.
Egress policies are based on

service providers MPLS EXP
markings.
Figure 15-15 illustrates adapting the ve-class service provider model to Pipe Mode. The
rst set of re-markings shows ingress PE re-marking from DSCP to MPLS EXP values,
depending on whether the trafc is in contract or out-of-contract. The second set of
markings shows how these MPLS EXP values can be mapped to QoS Groups (QG) and
Discard Classes (DC) to provide PHB classication and provisioning on PE-to-CE links
(without altering the IP DSCP values of the tunneled packets).
Example 15-11 shows the conguration for bidirectional re-marking on a PE router to
support Pipe Mode operation. Trafc received from CEs is marked explicitly (through
MPLS EXP values) to reect the service providers policies. Then trafc (traversing in the
opposite direction) received from the MPLS VPN core is mapped to QoS Groups and
Discard Classes so that PE-to-CE PHB egress policies can be performed against provider
re-markings. In this example, the customer has contracted for 3-Mbps service over an FE
link. Hierarchical policies are used to achieve queuing within (3 Mbps) shaping over this
(100-Mbps) link. Additionally, Discard-class WRED is enabled on the output queues so
that dropping decisions are based on Discard-class values (not IP ToS or DSCP values).
Furthermore, Discard-class dropping thresholds are tuned so that Discard-Class 1 (indicating
out-of-contract trafc) is dropped more aggressively than Discard-Class 0 (mimicking
DSCP-based WRED behavior), which is more consistent with RFC 2597 AssuredForwarding PHBs.
576
Figure 15-15 Five-Class Service Provider Model Pipe Mode Ingress and Egress Re-Marking Diagram
Ingress PE
MPLS VPN Cloud
Egress PE
EF
Real-Time
MPLS EXP 5
CS5
CS6
AF31
CS3
QG5
Real-Time
(Dropped)
Critical
Data
MPLS EXP 3
MPLS EXP 7
Video In Contract
AF21
CS2
Video
AF11
CS1
Bulk Data
Best-Effort
MPLS EXP 2
QG3
QG3/DC1
QG2
Critical
Data
Video
(Dropped)
MPLS EXP 1
MPLS EXP 6
QG1
Bulk Data
QG1/DC1
MPLS EXP 0
MPLS EXP 4
QG0
Best-Effort
QG0/DC1
Example 15-11 PE Conguration for MPLS DiffServ Pipe Mode Tunneling

!
match ip dscp ef
match ip dscp cs5
match ip dscp cs6
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp cs2
match ip dscp af11
match ip dscp cs1
!
!
class-map match-all MPLS-EXP-7
match mpls experimental topmost
! Matches MPLS EXP 7
577
Example 15-11 PE Conguration for MPLS DiffServ Pipe Mode Tunneling (Continued)
!
!
class-map match-all
match qos-group 5
class-map match-all
match qos-group 3
class-map match-all
match qos-group 2
class-map match-all
match qos-group 1
class-map match-all
match qos-group 0
QOSGROUP5
! Matches QoS Group 5
QOSGROUP3
QOSGROUP2
QOSGROUP1
QOSGROUP0
!
!
policy-map PIPE-MARKING
! Sets MPLS EXP Values
class REALTIME
police cir 1050000
conform-action set-mpls-exp-topmost-transmit 5 ! Conforming RT set to 5
exceed-action drop
class CRITICAL-DATA
police cir 600000
conform-action set-mpls-exp-topmost-transmit 3 ! Critical Data set to 3
exceed-action set-mpls-exp-topmost-transmit 7 ! Excess Critical set 7
class VIDEO
police cir 450000
conform-action set-mpls-exp-topmost-transmit 2 ! Conforming Video set to 2
exceed-action drop
class BULK-DATA
police cir 150000
conform-action set-mpls-exp-topmost-transmit 1 ! Conforming Bulk set to 1
exceed-action set-mpls-exp-topmost-transmit 6 ! Excess Bulk set to 6
class class-default
police cir 750000
conform-action set-mpls-exp-topmost-transmit 0 ! Conforming BE set to 0
!
!
policy-map MPLSEXP-QOSGROUP-DISCARDCLASS
! Maps MPLS EXP to QG/DC values
class MPLS-EXP-5
set qos-group 5
! Conforming Realtime is set to QG 5
class MPLS-EXP-3
set qos-group 3
! Conforming Critical Data is set to QG 3
continues
578
class MPLS-EXP-7
set qos-group 3
set discard-class 1
class MPLS-EXP-2
set qos-group 2
class MPLS-EXP-1
set qos-group 1
class MPLS-EXP-6
set qos-group 1
set discard-class 1
class MPLS-EXP-0
set qos-group 0
class MPLS-EXP-4
set qos-group 0
set discard-class 1
! Excess Critical Data is set to QG3

! Excess Critical Data has DC set to 1
! Conforming Video is set to QG 2
! Conforming Bulk is set to QG 1
! Excess Bulk is set to QG 1
! Excess Bulk has DC set to 1
! Conforming Best Effort is set to QG 0
! Excess Best Effort is set to QG 0
! Excess Best Effort has DC set to 1
!
!
policy-map PE-CE-QUEUING
! Queuing policy for PE to CE link
class QOSGROUP5
priority percent 35
! Voice class gets 35% LLQ
class QOSGROUP3
! Critical Data class gets 20% CBWFQ
random-detect discard-class-based
! DC-Based WRED is enabled
random-detect discard-class 0
30
40
10
! DC 0 is tuned for WRED
20
40
10
class QOSGROUP2
! Video class gets 15% CBWFQ
class QOSGROUP1
bandwidth percent 5
! Bulk class gets 5% CBWFQ
30
40
10
20
40
10
class QOSGROUP0
! Best Effort class gets 25% CBWFQ
30
40
10
20
40
10
!
!
policy-map PE-CE-SHAPING-QUEUING
class class-default
shape average 3000000
service-policy PE-CE-QUEUING
!
interface ATM2/0
no ip address
!
! Customer has 3 Mbps CIR over FE

! Shaping policy for 3 Mbps CIR
! Nested queuing policy
579
description ATM-OC3 TO MPLS VPN CORE
! Link to/from MPLS VPN Core
ip address 20.2.34.4 255.255.255.0
pvc 0/304
vbr-nrt 149760 149760
service-policy input MPLSEXP-QOSGROUP-DISCARDCLASS ! MPLS EXP to QG/DC
!
tag-switching ip
!
!
! Link to/from CE
ip address 10.1.12.2 255.255.255.0
! Pipe marking policy
service-policy input PIPE-MARKING
service-policy output PE-CE-SHAPING-QUEUING
! Shaping/Queuing policy
!
show policy
Pipe Mode with an Explicit Null LSP

When CEs are provider managed, some providers prefer to ofoad the ingress MPLS EXP
marking of customer trafc from the PE and push these policies out to the ingress interface
of the CE. However, because the CE-to-PE link is regular IP (not MPLS), a difculty arises
as to how to set the providers marking without affecting the IP DiffServ markings that the
customer has set (because, again, these are to be preserved and untouched through the
MPLS VPN cloud in Pipe Mode operation). Therefore, a solution to this scenario was
introduced in Cisco IOS Release 12.2(13)T, with the Pipe Mode MPLS DiffServ tunneling
with an Explicit Null LSP feature.
This feature prepends an Explicit Null LSP label for customer trafc headed from the CE
to the PE. This label is not used for MPLS switching; it is used only to preserve the providers MPLS EXP markings over the CE-to-PE link. On the PE, the MPLS EXP values are
copied to regular MPLS labels that are pushed onto the packet (which are used for MPLS
switching), and the explicit null label is discarded.
Thus, the ingress marking policies from the PE are pushed to the managed CE. This expands
the providers DiffServ domain to include the (managed) CEs. All other aspects of Pipe
Mode operation and conguration, however, remain the same.
580
Figure 15-16 MPLS DiffServ Pipe Mode with an Explicit Null LSP Tunneling Operation


are based on
provider markings.
MPLS VPN
CE Router
PE Router
PE Router
CE Router
P Routers
IPP3/DSCP AF31
Packet initially
marked to IPP3/
DSCO AF31.
MPLS EXP 4
IPP3/DSCP AF31
MPLS EXP 0
MPLS EXP 0
MPLS EXP 4
MPLS EXP values
IPP3/DSCP AF31
from IPP/DSCP values
on the managed CE
Topmost label is
via an explicit null LSP. marked down by a
policer.
IPP3/DSCP AF31
Topmost label is
label.
IPP3/DSCP AF31
values are
preserved.
Figure 15-17 shows the points where policies are required for Pipe Mode with Explicit Null
LSP MPLS DiffServ tunneling.
As noted, PE congurations remain the same as with normal Pipe Mode, with the exception
that ingress MPLS EXP marking policies have been removed. These policies now are set
on the managed CE, as shown in Example 15-12. As with Example 15-13, the provider has
contracted for a CIR of 3 Mbps in a ve-class model. All ingress trafc is policed on the
customer edge of the CE and is marked (through MPLS EXP values) to indicate whether it
is in contract or out-of-contract. Then an Explicit Null LSP is pushed onto the packet to
carry these MPLS EXP markings from the CE to the PE. Optionally, queuing policies can
be added on the CE-to-PE link, but for simplicity, these have been omitted from this
example because they already have been covered.
581
Figure 15-17 MPLS DiffServ Pipe Mode with an Explicit Null LSP Tunneling Policies
An explicit null LSP is

pushed onto the packet
with the providers MPLS
EXP markings.

group or Discard class on ingress
from the MPLS VPN cloud and then
the label is popped (and discarded).
MPLS VPN
(Ingress)
PE Router
CE Router
(Egress)
PE Router
CE Router
P Routers
MPLS EXP values are

explicitly set by
service provider.
Egress policies are based on

service providers MPLS EXP
markings.
Example 15-12 Managed CE Conguration for MPLS DiffServ Pipe Mode with an Explicit Null LSP
Tunneling
!
match ip dscp ef
match ip dscp cs5
match ip dscp cs6
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp cs2
match ip dscp af11
match ip dscp cs1
!
!
policy-map PIPE-EXPLICIT-NULL-MARKING
class REALTIME
police cir 1050000
exceed-action drop
! Conforming RT set to 5
continues
582
Example 15-12 Managed CE Conguration for MPLS DiffServ Pipe Mode with an Explicit Null LSP
Tunneling (Continued)
class CRITICAL-DATA
police cir 600000
class VIDEO
police cir 450000
exceed-action drop
class BULK-DATA
police cir 150000
class class-default
police cir 750000
!
!
description FE to Customer Network
ip address 10.1.1.1 255.255.255.0
service-policy input PIPE-EXPLICIT-NULL-MARKING
!
!
description FE TO PE
ip address 10.1.12.1 255.255.255.0
duplex auto
speed auto
mpls ip encapsulate explicit-null
!
! Critical Data set to 3

! Conforming Video set to 2

! Conforming Bulk set to 1

! Conforming BE set to 0
! Link to/from customer

! MPLS EXP set on ingress
! Link to/from PE
! Explicit Null LSP is added
show policy
Core QoS Considerations

Several options exist to meet strict SLA considerations for loss, delay, and jitter in the
service-provider MPLS VPN core:
Aggregate bandwidth overprovisioning

MPLS trafc engineering (which might or might not be used in conjunction with
DiffServ in the backbone)
583
Aggregate Bandwidth Overprovisioning

Aggregate bandwidth overprovisioning is a common trend in the service-provider
backbone because of its simplicity and ease of design, deployment, and operation.
Although DiffServ domain characteristics are assumed at the provider edges for trafc
aggregation, studies have shown that designing the service provider backbone for low
delay, jitter, or loss can be a matter of simply overprovisioning the network by
approximately two times the maximum of the aggregate trafc load.
Caveats to overprovisioning include capacity planning failures, network failure situations,
and unexpected trafc demands or patterns. For instance, because this strategy does not
differentiate between Real-Time trafc and Best-Effort trafc, real-time service levels can
be degraded if network failure or congestion occurs. Furthermore, this approach is
relatively expensive and inefcient, compared to alternative options such as Core DiffServ
and MPLS TE.
DiffServ in the Backbone

Deploying a simplied DiffServ policy in the backbone allows the service provider to
support multiple classes of trafc with different overprovisioning and underprovisioning
ratios on a per-class basis.
DiffServ in the backbone allows for more aggregate trafc to be supported for the same
provisioned network bandwidth, as compared to the non-DiffServ backbone. Also, RealTime trafc can be serviced preferentially, even if failure or congestion occurs.
The main caveat to this solution is that it adds complexity to the core network design and
operations.
In a DiffServ backbone, it might not be necessary to assume the same number of classes
that exist at the edge in the PE-to-CE link, as long as provision is made for a Real-Time
trafc class (which is associated to a PQ) and Critical Data is protected over BestEffort trafc.
Three-Class Provider-Core Model: PE-to-P or P-to-P Design

It is not necessary to ensure that the backbone supports the same number of DiffServ classes
as the edge, assuming that proper design principles are in place to support the given SLAs.
Figure 15-18 gives an example of how a ve-class provider-edge model can be collapsed
into a three-class provider-core model.
584
Figure 15-18 Five-Class Provider-Edge Model Short Pipe/Pipe Mode Mapping to Three-Class Provider-Core
Model Example
Edge DiffServ Classes
Core DiffServ Classes

EF
Real-Time
MPLS EXP 5
CS5
CS6
AF31
CS3
EXP5
(Dropped)
Critical
Data
MPLS EXP 3
MPLS EXP 7
EXP3
EXP7
Video In Contract
AF21
CS2
Video
MPLS EXP 2
EXP2
Bulk Data
MPLS EXP 1
MPLS EXP 6
EXP1
EXP6
Core
Critical
Data
(Dropped)
AF11
CS1
Core
Real-Time
Best-Effort
MPLS EXP 0
EXP0
MPLS EXP 4
Core
Best-Effort
EXP4
Detailed denitions of the core classes used in this gure follow:
Core Real-TimeThis class targets applications such as Voice and InteractiveVideo, which require low loss (less than 0.25 percent), low delay, and low jitter
(typically 5 ms within the backbone), and have a dened availability. This class also
supports per-ow sequence preservation. This class always should be engineered for
the worst-case delay, to support Real-Time trafc. Excess trafc in this class typically
is dropped. This class should be associated with MPLS EXP 5 with a PQ, to ensure
that the delay and jitter contracts are met. Between 25 and 35 percent of link capacity
should be allocated to the PQ. WRED typically is not required on this queue.
Core Critical DataThis class represents business-critical interactive applications.

Round-trip time (RTT) should be less than 250 ms (the threshold for human delay
perception), and loss should be less than 0.5 percent. Throughput is derived from loss
and RTT. Jitter is not important for this service class and is not dened. Excess trafc
within this class typically is re-marked with an out-of-contract identier (re-marking
of either DSCP or MPLS EXP values) and transmitted. This class also supports perow sequence preservation. This class should be associated with an AF class-based
queue or MPLS EXP-based queue (depending on the MPLS DiffServ tunneling mode
in use). WRED could be congured here to optimize TCP throughput and to
accommodate a drop policy for out-of-prole trafc.
585
Core Best-EffortThis class represents all other customer trafc that has not been
classied as Real-Time or Critical Data. It is dened in terms of a loss rate with
availability; throughput is derived from loss. Delay and jitter are not important for this
service and are not dened.
Example 15-13 shows the conguration of a three-class provider-core model. If DiffServ is

implemented within the core, the policies should be kept as simple as possible, to reduce
the CPU utilization for QoS to an absolute minimum.
Example 15-13 Three-Class Provider Core Model Example
!
class-map match-all CORE-REALTIME
match mpls experimental topmost 5
! Identifies in-contract Realtime
class-map match-all CORE-CRITICAL-DATA
! Identifies in-contract Critical-Data
! Identifies out-of-contract Critical Data
! Identifies in-contract Video
! Identifies in-contract Bulk
! Identifies out-of-contract Bulk
!
!
policy-map CORE-THREE-CLASS-SP-MODEL
class CORE-REALTIME
! CORE-REALTIME gets 35% LLQ
priority percent 35
class CORE-CRITICAL-DATA
! CORE-CRITICAL gets 55% CBWFQ
class class-default
! CORE-BEST-EFFORT gets FQ
fair-queue
!
show policy

policy is applied.
Platform-Specic ConsiderationsCisco 12000 GSR MDRR Example

Because of the highly specialized role of core routers, not all previously discussed QoS
features are supported by each platform/line card combination. It is beyond the scope of this
chapter to enumerate all platform/line card caveats; however, an example of the Cisco
12000 GSR can illustrate the point.
The Cisco 12000 series router platforms support queuing through a modied-decit roundrobin (MDRR) algorithm. MDRR works in much the same manner as CBWFQ and allows
for up to eight queues to be provisioned; non-empty queues are serviced in a round-robin
fashion, depending on the bandwidth assigned to them.
586
Within these eight queues, MDRR maintains a priority queue, which can be serviced in
strict priority or alternate priority.
With MDRR strict priority, the priority queue always is serviced exhaustively, as long as
there are packets in this queue. An important distinction between MDRR priority queuing
and LLQ is that MDRR priority queuing does not have an implicit policer. However, the
MQC syntax does support conguring MDRR strict priority in conjunction with classbased policing to enforce a limit on the amount of trafc that can be given strict priority.
To prevent starvation scenarios that could exist with the lack of a policer, an alternative to
MDRR strict-priority servicing exists in the option of MDRR alternate-priority queuing.
When alternate-priority queuing is congured, the priority queue is serviced alternately
with all other queues. For example, one priority packet is serviced, followed by a packet
from any one of the other queues (depending on how they are congured). Then another
priority packet is serviced and again is followed by a nonpriority packet. In this manner,
priority servicing does not starve other trafc as strict-priority queuing could.
MDRR can be applied to an inbound interface or an outbound interface.
Example 15-14 shows how to adapt the previous three-class provider-core policy to a Cisco
12000 GSR router. In this example, the LLQ is policed to approximately 35 percent of the
Packet-over-SONET (POS) line rate of OC3 (155 Mbps). Because the Cisco 12000 GSR
does not support WFQ, no explicit conguration is required for the Best-Effort class.
Example 15-14 Three-Class Provider Core Model Adapted for a Cisco 12000 GSR with Line Card 3
Example
!
match mpls experimental 5
!
!
policy-map CORE-THREE-CLASS-SP-MODEL-GSR
class CORE-REALTIME
! Voice is explicitly policed to 33%
police cir 54248000 bc 4470 be 4470
exceed-action drop
priority
! Voice is assigned strict-priority MDRR
!
! WFQ is not supported so no default-class polices are required
!
587
Example 15-14 Three-Class Provider Core Model Adapted for a Cisco 12000 GSR with Line Card 3
Example (Continued)
interface POS0/0
ip address 10.131.160.233 255.255.255.252
no ip directed-broadcast
tag-switching ip
crc 16
service-policy output CORE-THREE-CLASS-SP-MODEL-GSR
!
NOTE
The implementation of the priority command on the Cisco 12000 series differs from the
implementation on other routers running Cisco IOS Software. On this platform, the priority
trafc is not limited to the congured kbps value during periods of congestion. Thus, you
also must congure the police command to limit how much bandwidth a priority class can
use and to ensure adequate bandwidth for other classes. At this time, the police command
is supported on only Engine 3 line cards. On the other engine line cards, only class-default
is allowed when you congure a priority class.
show policy

Trafc engineering enables service providers to route network trafc efciently to offer the
optimal service levels, in terms of throughput and delay. Such effective use of network
resources allows providers to offer more granular services and, at the same time, reduce the
cost of their network.
Currently, many service providers base their services on an overlay model. In the overlay
model, transmission facilities are managed by Layer 2 switching. The routers see only a
fully meshed virtual topology, making most destinations appear one hop away. MPLS
trafc engineering (MPLS TE) provides a way to achieve the same trafc-engineering
benets of the overlay model without needing to run a separate network and without
needing a nonscalable, full mesh of router interconnects. Furthermore, MPLS TE enables
administrators to use complex metrics in determining the optimal path of a given ow.
MPLS trafc engineering does the following:
Replaces the need to manually congure the network devices to set up explicit routes.
Instead, you can rely on the MPLS trafc engineering functionality to understand the
backbone topology and the automated signaling process.
588
Accounts for link bandwidth and for the size of the trafc ow when determining
explicit routes across the backbone.
Has a dynamic adaptation mechanism that enables the backbone to be resilient to

failures, even if several primary paths are precalculated ofine.
MPLS TE automatically establishes and maintains a tunnel across the backbone, using
RSVP. The path used by a given tunnel at any point in time is determined based on the
tunnel resource requirements and network resources, such as bandwidth.
Available resources are ooded through extensions to a link statebased Interior Gateway
Protocol (IGP), such as OSPF or IS-IS.
Tunnel paths are calculated at the tunnel head based on a t between required and available
resources (constraint-based routing). The IGP automatically routes the trafc into these
tunnels. Typically, a packet crossing the MPLS TE backbone travels on a single tunnel that
connects the ingress point to the egress point.
MPLS TE is built on the following Cisco IOS mechanisms:
Label-switched path (LSP) tunnels, which are signaled through RSVP, with trafc
engineering extensions. LSP tunnels are represented as tunnel interfaces, have a
congured destination, and are unidirectional.
A link-state IGP (such as OSPF or IS-IS) with extensions for the global ooding of
resource information, and extensions for the automatic routing of trafc onto LSP
tunnels, as appropriate.
An MPLS TE path-calculation module that determines paths to use for LSP tunnels.
An MPLS TE link-management module that does link admission and bookkeeping of
the resource information to be ooded.
One approach to engineer a backbone is to dene a mesh of tunnels from every ingress
device to every egress device. The IGP, operating at an ingress device, determines which
trafc should go to which egress device, and steers that trafc into the tunnel from ingress
to egress. The MPLS trafc-engineering path-calculation and signaling modules determine
the path taken by the LSP tunnel, subject to resource availability and the dynamic state of
the network. For each tunnel, counts of packets and bytes sent are kept.
Sometimes a ow is so large that it cannot t over a single link, so it cannot be carried by
a single tunnel. In this case, multiple tunnels between a given ingress and egress can be
congured, and the ow is load-shared among them.
Basic MPLS TE
Basic MPLS TE conguration requires you to do the following:
Enable IP CEF globally:

Router(config)#ip
ip cef
589
Enable MPLS TE globally:
Enable MPLS TE on every interface that might be part of a tunnel, and congure the
amount of bandwidth that can be reserved (through RSVP) for all tunnels traversing
the link:
Router(config)#mpls
mpls traffic-eng tunnels
Router(config)#interface POS5/0
Router(config-if)#mpls
Router(config-if)#ip
Dene a loopback interface to be used as an MPLS TE Router ID (RID):

Router(config)#interface
interface loopback 0
Router(config-if)#ip address 20.1.1.1 255.255.255.255
Congure a link-state Interior Gateway Protocol (IGP) for MPLS TE, either OSPF or
IS-IS:
OSPF
Router(config)#router
router ospf 100
Router(config-router)# mpls traffic-eng router-id loopback 0
Router(config-router)# mpls traffic-eng area 0
IS-IS
Router(config)#router
router isis
Router(config-router)#net 49.0000.0000.0000.0010.00
Router(config-router)#metric-style
metric-style wide
Router(config-router)#mpls
mpls traffic-eng level-1
Router(config-router)#mpls
mpls traffic-eng router-id loopback 0
Dene a tunnel interface congured with MPLS TE, with an amount of bandwidth
that can be reserved (through RSVP) for the tunnel:
Router(config)#interface
interface Tunnel0
Router(config-if)# description BLUE-TUNNEL (PE1=>PE2)
Router(config-if)# ip unnumbered loopback 0
Router(config-if)# tunnel destination 20.2.2.2
Router(config-if)# tunnel mode mpls traffic-eng
Router(config-if)# tunnel mpls traffic-eng bandwidth 77500
Set a priority for the tunnel interface (lower is preferred):
Select a path option for the tunneldynamic, explicit, or a combination:
Router(config-if)# tunnel mpls traffic-eng priority 7 7
Dynamic path optionThe software calculates the best path to the

destination, based on tunnel constraints.
Router(config-if)#tunnel
tunnel mpls traffic-eng path-option 10 dynamic
Explicit path optionEach hop is dened statically.

Router(config-if)# tunnel mpls
mpls traffic-eng
traffic-eng path-option
path-option 1 explicit name BLUE-TUNNEL
Router(config-if)#exit
Router(config)#ip
ip explicit-path name BLUE-TUNNEL enable
Router(cfg-ip-expl-path)# next-address 20.1.12.2
Explicit Path name BLUE-TUNNEL:
1: next-address 20.1.12.2
Router(cfg-ip-expl-path)#
590
Combination path optionsWhen the rst option cannot be met, the

algorithm attempts the second, and so on.
Router(config-if)# tunnel mpls traffic-eng path-option 1 explicit name
TUNNEL
Router(config-if)# tunnel mpls traffic-eng path-option 2 dynamic
Router(config)#ip
Router(cfg-ip-expl-path)# next-address 20.1.12.2
Explicit Path name BLUE-TUNNEL:
1: next-address 20.1.12.2
Router(cfg-ip-expl-path)#
BLUE-
Select a forwarding option to direct trafc down the tunnelstatic routes, policy
routing, or AutoRoute:
Static routes
Router(config)#ip
ip route 16.16.16.16 255.255.255.255 Tunnel0
Policy-based routing
Router(config)#interface FastEthernet 1/0
Router(config-if)#ip
ip policy route-map TUNNEL-ASSIGNMENT
Router(config)#route-map
route-map TUNNEL-ASSIGNMENT
Router(config-route-map)#match
match ip address 100
Router(config-route-map)#set
set interface Tunnel 0
Router(config-route-map)#exit
Router(config)#access-list 100 permit ip any 10.2.2.0 0.0.0.255
AutoRoute (enabled on the tunnel headend)

Router(config)#interface Tunnel0
Router(config-if)#tunnel
tunnel mpls traffic-eng autoroute announce
NOTE
As stated at the opening of the chapter, it is simply beyond the scope of this book to go into
detail on MPLS trafc engineering. Many MPLS TE features, such as attribute ags,
afnity, administrative weights, ooding thresholds, and FastReroute, have been omitted
from this overview. For a comprehensive technical discussion on these and other MPLS TE
features and operation, refer to the Cisco Press book Trafc Engineering with MPLS, by
Eric Osborne and Ajay Simha.
Two examples of QoS-related MPLS TE scenarios are presented in this chapter. The rst is
the use of MPLS per-VPN trafc engineering; the second is the use of MPLS DiffServ
trafc engineering (MPLS DS-TE).
591
MPLS Per-VPN TE
MPLS VPNs provide the route isolation necessary for data privacy and support of private
IP address allocation overlapping, while RSVP-signaled MPLS-TE provides optimal path
selection, per-VPN SLAs, and path resiliency.
The notion of integrating these two MPLS applications is pretty straightforward. Where it
gets interesting is in the actual binding process of an MPLS VPN to the TE tunnel. Today
this can be achieved by taking advantage of next-hop route recursion at the PE; however,
this is a less than elegant interim solution. Cisco IOS innovations in policy-based routing
(PBR) and QoS-based routing (QBR) will allow for a more scalable way to integrate MPLS
VPN and TE in the near future.
In this simplied example scenario, the service provider has Packet over SONET (POS)
links from each PE to the core, and also direct PE-to-PE POS links for geographically
adjacent PEs. Furthermore, the service provider has two customers, Blue and Red.
Customer Blue has contracted with the provider for a premium service so that Blue trafc
that is destined for geographically adjacent CEs is tunneled over the direct PE-to-PE links.
However, Red trafc always is switched through the MPLS core, even when there is a
shorter path to the destination CEs through the direct PE-to-PE links. Figure 15-19
illustrates the desired operation of this scenario. Because MPLS TE is unidirectional,
separate tunnels would be required for the same behavior in the return direction.
Figure 15-19 MPLS Per-VPN TE Operation Example
Customer Blue traffic is sent over the direct
link to the adjacent PE (not through core).
Blue CE Router
(Ingress)
PE Router
(Egress)
PE Router
Red CE Router
Customer Red traffic

is always sent
through the core.
Red CE Router
P Routers
MPLS VPN
Core
Blue CE Router
592
Figure 15-20 zooms in on the relevant subsection of the previous diagram and provides
specic addressing detail for this example.
Figure 15-20 MPLS Per-VPN TE Example Details
Blue CE1 Router
10.1.1.0/24
10.20.1.0/30
PE1
PE2
Tunnel 0
Blue CE2 Router

10.2.2.0/24
10.20.2.0/30
20.1.12.0/30
15.1.1.0/24
15.20.1.0/30
15.2.2.0/24
Lo: 20.2.2.2
Lo: 20.1.1.1/32
Red CE1 Router
Red CE2 Router
20
.1
.2
3.
0/
30
30
0/
3.
.1
.1
20
Customer Red traffic

is always sent
through the core.
15.20.2.0/30
Tunnel 1
Lo: 20.3.3.3
P Router
MPLS VPN Core
The conguration for this example of MPLS per-VPN TE spans three routers: PE1 conguration (see Example 15-15), PE2 conguration (see Example 15-16), and the P-router conguration (see Example 15-17).
To achieve the desired behavior of forcing Customer Blues trafc down the shorter MPLS
tunnel and forcing Customer Reds trafc down the longer tunnel, a combination of static
and BGP policy-based routing is used (relying on recursive routing on each PE).
When Customer Blues geographically adjacent CE networks are received at the PE, BGP
directs these prexes to be processed by the TUNNEL-ASSIGNMENT route map. This
route map matches Blues prexes against access-list 1 and sets the BGP next-hop attribute
to the ctitious route of 16.16.16.16. Similarly, Customer Reds prexes are matched
against access-list 2 and have their BGP next-hop attribute set to the ctitious route of
17.17.17.17. Then each of these ctitious routes is resolved through static-routing
statements that direct the ow to the desired MPLS tunnel interface. This is detailed in
Example 15-15.
Example 15-15 PE1 MPLS Per-VPN TE Example Conguration
!
hostname PE1
!
593
Example 15-15 PE1 MPLS Per-VPN TE Example Conguration (Continued)

!
ip vrf BLUE
! BLUE MPLS VPN definition
rd 100:1
route-target export 100:1
route-target import 100:1
!
ip vrf RED
! RED MPLS VPN definition
rd 150:1
!
ip cef
! IP CEF is required for MPLS and MPLS TE
mpls ldp logging neighbor-changes
! MPLS TE is enabled globally
!
!
interface Loopback0
! Loopback interface for MPLS TE RID
ip address 20.1.1.1 255.255.255.255
!
interface Tunnel0
description BLUE-TUNNEL (PE1=>PE2)
ip unnumbered Loopback0
tunnel destination 20.2.2.2
! MPLS TE is enabled on Tunnel int
tunnel mode mpls traffic-eng
! Lower value is preferred
tunnel mpls traffic-eng priority 7 7
! 77.5 Mbps can be reserved
tunnel mpls traffic-eng bandwidth 77500
tunnel mpls traffic-eng path-option 1 explicit name BLUE-TUNNEL ! BLUE Path
!
interface Tunnel1
description RED-TUNNEL (PE1=>P=>PE2)
! 77.5 Mbps can be reserved via RSVP
! RED Path
tunnel mpls traffic-eng path-option 1 explicit name RED-TUNNEL
!
!
description FE to Customer Blue CE1
ip vrf forwarding BLUE
! Interface to Blue CE1
ip address 10.20.1.2 255.255.255.252
!
description FE to Customer Red CE1
ip address 15.20.1.2 255.255.255.252
!
interface POS5/0
ip address 20.1.12.1 255.255.255.252
description PE1=>PE2 POS Link
continues
594

tag-switching ip
! MPLS TE enabled on int

! MPLS enabled on int
! RSVP enabled on int for 77.5 Mbps
!
interface POS6/0
description PE1=>P-Router (Core) POS Link
ip address 20.1.13.1 255.255.255.252
tag-switching ip
!
router ospf 100
mpls traffic-eng router-id Loopback0
! MPLS TE RID
mpls traffic-eng area 0
! Enables OSPF area 0 for MPLS TE
log-adjacency-changes
redistribute connected subnets
network 20.1.12.0 0.0.0.3 area 0
network 20.1.13.0 0.0.0.3 area 0
!
router bgp 100
no synchronization
bgp log-neighbor-changes
redistribute connected
neighbor 20.2.2.2 remote-as 100
neighbor 20.2.2.2 update-source Loopback0
no auto-summary
!
address-family vpnv4
neighbor 20.2.2.2 activate
! MPLS VPN neighbor (PE2)
neighbor 20.2.2.2 send-community extended
neighbor 20.2.2.2 route-map TUNNEL-ASSIGNMENT in ! Applies BGP PBR
exit-address-family
!
address-family ipv4 vrf BLUE
! EBGP peer with Customer Blue CE1
neighbor 10.20.1.1 default-originate
no auto-summary
no synchronization
exit-address-family
!
address-family ipv4 vrf RED
! EBGP peer with Customer Red CE1
no auto-summary
no synchronization
exit-address-family
595

!
ip classless
ip route 16.16.16.16 255.255.255.255 Tunnel0
ip route 17.17.17.17 255.255.255.255 Tunnel1
!
ip bgp-community new-format
!
next-address 20.1.12.2
!
ip explicit-path name RED-TUNNEL enable
!
access-list 1 permit 10.2.2.0 0.0.0.255
!
route-map TUNNEL-ASSIGNMENT permit 10
match ip address 1
set ip next-hop 16.16.16.16
!
match ip address 2
!
!
! Static Route for BLUE Tunnel

! Static Route for RED Tunnel
! Explicit Path for BLUE Tunnel

! Direct to PE2
! Explicit Path for RED Tunnel
! First to P-Router (Core)
! Then to PE2
!
!
!
!
Adjacent
Adjacent
Adjacent
Adjacent
Customer
Customer
Customer
Customer
Blue subnets
Blue subnets
Red subnets
Red subnets
! BGP Inbound Route-Map

! Identifies Customer Blue subnets
! Sets BGP Next-Hop to 16.16.16.16
! Identifies Customer Red subnets

Because all tunnels are unidirectional, the same logic is applied in the reverse direction,
where the ctitious addresses are 18.18.18.18 (for Blue) and 19.19.19.19 (for Red), as
detailed in Example 15-16.
Example 15-16 PE2 MPLS Per-VPN TE Example Conguration
!
hostname PE2
!
!
ip vrf BLUE
rd 100:1
route-target
route-target
!
ip vrf RED
rd 150:1
route-target
route-target
!

export 100:1
import 100:1
export 150:1
import 150:1
continues
596

ip cef
!
!
interface Loopback0
ip address 20.2.2.2 255.255.255.255
!
interface Tunnel0
description BLUE-TUNNEL (PE2=>PE1)
! 77.5 Mbps can be reserved
tunnel mpls traffic-eng path-option 1 explicit name BLUE-TUNNEL ! BLUE Path
!
interface Tunnel1
description RED-TUNNEL (PE2=>P=>PE1)
! 77.5 Mbps can be reserved via RSVP
tunnel mpls traffic-eng path-option 1 explicit name RED-TUNNEL
! RED Path
!
!
ip address 10.20.2.2 255.255.255.252
!
! Interface to Red CE2
ip address 15.20.2.2 255.255.255.252
!
interface POS5/0
ip address 20.1.12.2 255.255.255.252
tag-switching ip
!
interface POS6/0
ip address 20.1.23.1 255.255.255.252
tag-switching ip
!
router ospf 100
597

network 20.1.12.0 0.0.0.3 area 0
network 20.1.23.0 0.0.0.3 area 0
! MPLS TE RID
!
router bgp 100
no synchronization
no auto-summary
!
! MPLS VPN neighbor (PE1)
neighbor 20.1.1.1 route-map TUNNEL-ASSIGNMENT in
! Applies BGP PBR
exit-address-family
!
! EBGP peer with Customer Blue CE2
no auto-summary
no synchronization
exit-address-family
!
! EBGP peer with Customer Red CE2
no auto-summary
no synchronization
exit-address-family
!
ip classless
ip route 18.18.18.18 255.255.255.255 Tunnel0 ! Static Route for BLUE Tunnel
ip route 19.19.19.19 255.255.255.255 Tunnel1 ! Static Route for RED Tunnel
!
!
! Explicit Path for BLUE Tunnel
! Direct to PE1
!
ip explicit-path name RED-TUNNEL enable
! Explicit Path for RED Tunnel
! First to P-Router (Core)
! Then to PE1
continues
598

!
!
match ip address 1
!
match ip address 2
!
!
!
!
!
Adjacent
Adjacent
Adjacent
Adjacent
Customer
Customer
Customer
Customer
Blue subnets
Blue subnets
Red subnets
Red subnets
! BGP Inbound Route-Map

! Identifies Customer Blue subnets
! Identifies Customer Red subnets

A sample P-router conguration is shown in Figure 15-17 to illustrate the MPLS TE

commands required in the core.
Example 15-17 P-Router MPLS Per-VPN TE Example Conguration
!
hostname P-Router
!
!
ip cef
!
!
!
interface Loopback0
ip address 20.3.3.3 255.255.255.255
!
!
interface POS5/0
description P-Router (Core) => PE1 POS Link
ip address 20.1.13.2 255.255.255.252
tag-switching ip
!
interface POS6/0
ip address 20.1.23.2 255.255.255.252
tag-switching ip
599
Example 15-17 P-Router MPLS Per-VPN TE Example Conguration (Continued)

!
router ospf 100
network 20.1.13.0 0.0.0.3 area 0
network 20.1.23.0 0.0.0.3 area 0
!
! MPLS TE RID
show ip rsvp interface

show ip rsvp neighbor
show mpls interface
show mpls trafc-eng tunnels summary
show mpls trafc-eng tunnels show mpls trafc-eng topology
show ip bgp vpnv4 all
ping vrf with show interface tunnel
Verication Command: show ip rsvp interface

The show ip rsvp interface command is useful to verify whether all physical interfaces that
can participate in tunnels have been congured with an amount of reservable bandwidth. If
the command ip rsvp bandwidth is omitted from the interface conguration, the amount
of reservable bandwidth for that interface defaults to 0. A common cause of tunnels not
coming up is the oversight of conguring RSVP reservations on the physical interface. See
Example 15-18.
Example 15-18 Verication of RSVP Conguration on Local Router
PE1#show ip rsvp interface
interface
allocated i/f max
PO5/0
77500K
77500K
PO6/0
77500K
77500K
PE1#
flow max sub max

77500K
0
77500K
0
Verication Command: show ip rsvp neighbor

A similar command for checking whether RSVP has been enabled properly on physical
interfaces is the show ip rsvp neighbor command, which reports whether RSVP has been
enabled on neighboring router interfaces. See Example 15-19.
600
Example 15-19 Verication of RSVP Conguration on Neighboring Routers

PE1#show ip rsvp neighbor
20.1.12.2
RSVP
20.1.13.2
RSVP
PE1#
Verication Command: show mpls interface

The show mpls interface command is a useful command to verify whether tag switching
has been congured properly on local interfaces. This command also lists which interfaces
may participate in tunnels. Notice that tag switching is not congured explicitly on tunnel
interfaces, although they are advertised through the link-state IGP. See Example 15-20.
Example 15-20 Verication of Tag-Switching Conguration
PE1#show mpls interfaces
Interface
IP
POS5/0
Yes (tdp)
POS6/0
Yes (tdp)
Tunnel0
No
Tunnel1
No
PE1#
Tunnel
Yes
Yes
No
No
Operational
Yes
Yes
Yes
Yes
Verication Command: show mpls trafc-eng tunnels summary

A quick way to check whether MPLS TE has been enabled globally and has been
congured properly is to use the show mpls trafc-eng tunnels summary command.
Example 15-21 shows an example of this command: Two tunnel heads and two tunnel tails
have been congured correctly and are active.
Example 15-21 Verication of MPLS TE Global Conguration
PE1#show mpls traffic-eng tunnels summary
Signalling Summary:
LSP Tunnels Process:
running
RSVP Process:
running
Forwarding:
enabled
Head: 2 interfaces, 2 active signalling attempts, 2 established
3 activations, 1 deactivations
Midpoints: 0, Tails: 2
Periodic reoptimization:
every 3600 seconds, next in 1712 seconds
Periodic auto-bw collection:
disabled
PE1#
Verication Command: show mpls trafc-eng tunnels

Additional per-tunnel detail is shown by the show mpls trafc-eng tunnels command. See
Example 15-22.
601
Example 15-22 Verication of MPLS Tunnel Details

PE1#show mpls traffic-eng tunnels tunnel 0
Name: BLUE-TUNNEL (PE1=>PE2)
(Tunnel0) Destination: 20.2.2.2
Status:
Status
Admin: up
Oper: up
Path: valid
Signalling: connected
path option 1, type explicit BLUE-TUNNEL (Basis for Setup, path weight 1)
Config Parameters:
Bandwidth: 77500
kbps (Global) Priority: 7 7
Affinity: 0x0/0xFFFF
Metric Type: TE (default)
AutoRoute: disabled LockDown: disabled Loadshare: 77500
bw-based
auto-bw: disabled
InLabel : OutLabel : POS5/0, implicit-null
RSVP Signalling Info:
Src 20.1.1.1, Dst 20.2.2.2, Tun_Id 0, Tun_Instance 7
RSVP Path Info:
My Address: 20.1.1.1
Explicit Route: 20.1.12.2 20.2.2.2
Record Route: NONE
Tspec: ave rate=77500 kbits, burst=1000 bytes, peak rate=77500 kbits
RSVP Resv Info:
Record Route: NONE
Fspec: ave rate=77500 kbits, burst=1000 bytes, peak rate=77500 kbits
Shortest Unconstrained Path Info:
Path Weight: 1 (TE)
Explicit Route: 20.1.12.2 20.2.2.2
History:
Tunnel:
Time since created: 2 hours, 34 minutes
Time since path change: 2 hours, 34 minutes
Current LSP:
Uptime: 2 hours, 34 minutes
PE1#
Verication Command: ping vrf with show interface tunnel

Verication of the MPLS TE solution can be performed with a ping vrf command coupled
with a show interface tunnel command to demonstrate that trafc for each VPN is being
sent over the proper tunnel. The interface counters can be cleared with the clear counters
command before these tests. See Example 15-23.
Example 15-23 Solution Verication with ping vrf and show interface tunnel Commands
PE1#ping vrf BLUE 10.2.2.2
Type escape sequence to abort.
Sending 5, 100-byte ICMP Echos to 10.2.2.2, timeout is 2 seconds:
!!!!!
Success rate is 100 percent (5/5), round-trip min/avg/max = 1/2/4 ms
PE1#ping vrf RED 15.2.2.2
continues
602
Example 15-23 Solution Verication with ping vrf and show interface tunnel Commands (Continued)
!!!!!
!!!!!
!!!!!
PE1#
PE1#show interface tunnel 0
Tunnel0 is up, line protocol is up
Hardware is Tunnel
Description: BLUE-TUNNEL (PE1=>PE2)
Interface is unnumbered. Using address of Loopback0 (20.1.1.1)
Encapsulation TUNNEL, loopback not set
Keepalive not set
Tunnel source UNKNOWN, destination 20.2.2.2
Tunnel protocol/transport Label Switching, key disabled, sequencing disabled
Checksumming of packets disabled, fast tunneling enabled
Last input never, output never, output hang never
Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops: 0
Queueing strategy: fifo
Output queue: 0/0 (size/max)
5 minute input rate 0 bits/sec, 0 packets/sec
5 minute output rate 0 bits/sec, 0 packets/sec
0 packets input, 0 bytes, 0 no buffer
Received 0 broadcasts, 0 runts, 0 giants, 0 throttles
0 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored, 0 abort
5 packets output,
output 500 bytes, 0 underruns
0 output errors, 0 collisions, 0 interface resets
0 output buffer failures, 0 output buffers swapped out
PE1#
PE1#show interface tunnel 1
Tunnel1 is up, line protocol is up
Hardware is Tunnel
Description: RED-TUNNEL (PE1=>P=>PE2)
Interface is unnumbered. Using address of Loopback0 (20.1.1.1)
Encapsulation TUNNEL, loopback not set
Keepalive not set
Tunnel source UNKNOWN, destination 20.2.2.2
Tunnel protocol/transport Label Switching, key disabled, sequencing disabled
Checksumming of packets disabled, fast tunneling enabled
603
Example 15-23 Solution Verication with ping vrf and show interface tunnel Commands (Continued)
Last input never, output never, output hang never
Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops: 0
Queueing strategy: fifo
Output queue: 0/0 (size/max)
5 minute input rate 0 bits/sec, 0 packets/sec
5 minute output rate 0 bits/sec, 0 packets/sec
0 packets input, 0 bytes, 0 no buffer
Received 0 broadcasts, 0 runts, 0 giants, 0 throttles
0 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored, 0 abort
15 packets output,
output 1500 bytes, 0 underruns
0 output errors, 0 collisions, 0 interface resets
0 output buffer failures, 0 output buffers swapped out
PE1#
MPLS DS-TE
MPLS trafc engineering allows constraint-based routing (CBR) of IP trafc. One of the
constraints satised by CBR is the availability of required bandwidth over a selected path.
DiffServ-aware trafc engineering extends MPLS trafc engineering to enable you to
perform constraint-based routing of guaranteed trafc, which satises a more restrictive
bandwidth constraint than that required for regular trafc. The more restrictive bandwidth
is termed a subpool, and the regular TE tunnel bandwidth is called the global pool (the
subpool is a portion of the global pool). This capability to satisfy a more restrictive
bandwidth constraint translates into a capability to achieve higher quality of service
performance (in terms of delay, jitter, or loss) for the guaranteed trafc.
For example, DS-TE can be used to ensure that trafc is routed over the network so that,
on every link, there is never more than 35 percent (or any assigned percentage) of the link
capacity of guaranteed trafc (for example, voice), while there can be up to 100 percent of
the link capacity of regular trafc. Assuming that QoS mechanisms also are used on every
link to queue guaranteed trafc separately from regular trafc, it then becomes possible to
enforce separate overbooking ratios for guaranteed and regular trafc.
Also, through the capability to enforce a maximum percentage of guaranteed trafc on
any link, the network administrator directly can control the end-to-end QoS performance
parameters without having to rely on overengineering or on expected shortest-path routing
behavior. This is essential for transport of applications that have very high QoS requirements (such as real-time voice, virtual IP leased line, and bandwidth trading), where
overengineering cannot be assumed everywhere in the network.
DS-TE involves extending the IGP link-state protocols so that the available subpool bandwidth at each preemption level is advertised in addition to the available global pool bandwidth
at each preemption level. DS-TE modies constraint-based routing to take this more complex advertised information into account during path computation.
604
DiffServ-aware trafc engineering enables service providers to perform separate admission

control and separate route computation for discrete subsets of trafc (for example, voice
and data trafc). Remember, as with expedited forwarding, PHBs can protect voice from
data, but only admission-control mechanisms can protect voice from voice.
For example, a service provider might sell voice services to two customers, both with
a strict-priority servicing requirement. However, if congestion is occurring on a node to
the point that not enough bandwidth is available for both customers voice trafc, MPLS
DS-TE can signal this condition back to the headend so that additional voice trafc can
be rerouted along a different path where adequate priority bandwidth still exists.
In short, MPLS DS-TE offers additional granularity and guarantees to trafc engineering
scenarios, which are suited especially to Real-Time applications such as Voice and
Interactive-Video.
These incremental steps are required for enabling MPLS DS-TE:
Provision per-link DiffServ policies:

PE1(config)#class-map match-all CORE-REALTIME
PE1(config-cmap)#match mpls experimental topmost 5
PE1(config-cmap)#class-map match-all CORE-CRITICAL-DATA
PE1(config-cmap)#policy-map CORE-THREE-CLASS-SP-MODEL
PE1(config-pmap)#class CORE-REALTIME
PE1(config-pmap-c)#priority
priority percent 35
PE1(config-pmap-c)#class CORE-CRITICAL-DATA
PE1(config-pmap-c)#bandwidth percent 55
PE1(config-pmap-c)#class class-default
PE1(config-pmap-c)#interface pos 5/0
PE1(config-if)# max-reserved-bandwidth 100
PE1(config-if)# service-policy output CORE-THREE-CLASS-SP-MODEL
PE1(config)#interface pos 6/0
PE1(config-if)# max-reserved-bandwidth 100
PE1(config-if)# service-policy output CORE-THREE-CLASS-SP-MODEL
PE1(config-if)#exit
Assign per-link reservable subpool bandwidth allotments on all physical interfaces

that are participating in DS-TE tunnels (including links on midpoint routers):
PE1(config)#interface pos 5/0
PE1(config-if)# ip rsvp bandwidth 77500 sub-pool 54250
Assign a reservable subpool bandwidth to each MPLS DS-TE tunnel headend:

PE1(config)#interface tunnel 0
PE1(config-if)#tunnel
tunnel mpls traffic-eng bandwidth sub-pool 54250
Lower the priority (remember lower is better) of the MPLS DS-TE tunnel so that it
can preempt reservations, as needed:
PE1(config)#interface tunnel 0
PE1(config-if)# tunnel mpls traffic-eng priority 0 0
Tighten down on the MPLS DS-TE headend tunnel admission control criteria (via
static routes and/or policy-based routing).
605
A simplifying assumption was made in the previous scenario (MPLS per-VPN TE) that
each customers subnets would not overlap. In the real world, however, this is an unlikely
assumption. Therefore, in this second MPLS TE example, additional ratcheting has been
done on the tunnel headends to allow only Voice trafc from Customer Blue to traverse the
shorter tunnel, despite the fact that, in this example, both Customer Blue and Customer Red
are using the same private address space. Voice trafc is identied as destined for the voice
VLAN subnets of the CE routers (10.1.10x.0/24), as shown in Figure 15-21.
Figure 15-21 MPLS DS-TE Example Details
10.1.1.0/24
10.1.101.0/24
Blue CE1 Router

10.20.1.0/30
Tunnel 0
PE1
PE2
Blue CE2 Router

10.20.2.0/30
20.1.12.0/30
10.20.1.0/30
Lo: 20.1.1.1/32
Lo: 20.2.2.2
Core DiffServ Policies:

Three-Class Per-Link
Scheduling Policies
Red CE1 Router
0/
20
30
.1
0/
.2
3.
3.
.1
Customer Blue data traffic and

all Customer Red traffic (both
voice and data) are always sent
through the core.
10.2.2.0/24
10.2.102.0/24
Red CE2 Router
.1
20
Tunnel 1
10.20.2.0/30
10.2.2.0/24
10.2.102.0/24
30
10.1.1.0/24
10.1.101.0/24
Lo: 20.3.3.3
P Router
MPLS VPN Core
In addition to identifying Voice trafc by destination subnet address, the BGP VPNv4 route
targets must match Customer Blues assigned route target (100:1). In this manner, only
Customer Blue, which has purchased a premium service, gains access to the direct DS-TE
tunnel to geographically adjacent CEs (despite the fact that Customer Red is using identical
private IP addresses).
NOTE
Only standard IP Extended Community lists (numbered 1 to 99) support the route target
(RT) keyword and functionality.
Additionally, very basic core DiffServ policies are applied to the POS links connecting PE
and P routers. A simplifying assumption is made that trafc has been marked correctly on
606
the CEs and is being mapped to MPLS EXP values using the standard (default) DSCPto-EXP mapping (for example, DSCP EF is mapped to MPLS EXP 5). For simplicity, no
policers are shown in Example 15-24; these are reintroduced, however, in the nal case
study example at the end of this chapter.
As with the previous example, the conguration for this MPLS DS-TE example spans three
routers: PE1 conguration (see Example 15-24), PE2 conguration (see Example 15-25),
and the P router (see Example 15-26) conguration.
Example 15-24 PE1 MPLS DS-TE Example Conguration
!
hostname PE1
!
!
ip vrf BLUE
rd 100:1
!
ip vrf RED
rd 150:1
!
ip cef
!
!
!
!
! Per-link scheduling policy
class CORE-REALTIME
priority percent 35
class class-default
fair-queue
!
!
interface Loopback0
ip address 20.1.1.1 255.255.255.255
!
interface Tunnel0
description TUNNEL0 (PE1=>PE2)
607
Example 15-24 PE1 MPLS DS-TE Example Conguration (Continued)

tunnel
tunnel
tunnel
tunnel
mode
mpls
mpls
mpls
mpls traffic-eng
traffic-eng priority 0 0
traffic-eng bandwidth sub-pool 54250
! MPLS DS-TE sub-pool
traffic-eng path-option 1 explicit name TUNNEL0
!
interface Tunnel1
description TUNNEL0 (PE1=>P=>PE2)
tunnel mpls traffic-eng path-option 1 explicit name TUNNEL1
!
!
ip address 10.20.1.2 255.255.255.252
! Same IP address as Red
!
ip address 10.20.1.2 255.255.255.252
! Same IP address as Blue
!
interface POS5/0
ip address 20.1.12.1 255.255.255.252
service-policy output CORE-THREE-CLASS-SP-MODEL
! Scheduling applied
tag-switching ip
ip rsvp bandwidth 77500 sub-pool 54250
! Sub-pool defined on int
!
interface POS6/0
ip address 20.1.13.1 255.255.255.252
tag-switching ip
! No change to PE-P RSVP
!
router ospf 100
network 20.1.12.0 0.0.0.3 area 0
network 20.1.13.0 0.0.0.3 area 0
continues
608

!
router bgp 100
no synchronization
no auto-summary
!
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
ip classless
ip route 16.16.16.16 255.255.255.255 Tunnel0 ! Static route for Tunnel 0
!
ip extcommunity-list 1 permit rt 100:1
! Identifies Blue VPN by RT
! Identifies Red VPN by RT
!
ip explicit-path name TUNNEL0 enable
!
!
! Identifies (Blue) Voice-VLAN
! Identifies (Blue) Data-VLAN
! Identifies (Blue) PE-CE link
! Identifies (Red) Voice-VLAN
609

!
match ip address 1
match extcommunity 1
!
match ip address 2
!
match ip address 3
!
! Identifies (Red) Data-VLAN

! Identifies (Red) PE-CE Link
! Matches Voice-VLAN subnet

! Matches Blue VPN RT
! Matches other (Blue) subnets

! Matches all (Red) subnets

! Matches Red VPN RT
The second/complementary PE routers conguration for this MPLS DS-TE example is

Example 15-25 PE2 MPLS DS-TE Example Conguration
!
hostname PE2
!
!
ip vrf BLUE
rd 100:1
!
ip vrf RED
rd 150:1
!
ip cef
!
!
!
continues
610

!
class CORE-REALTIME
priority percent 35
class class-default
fair-queue
!
!
interface Loopback0
ip address 20.2.2.2 255.255.255.255
!
interface Tunnel0
! MPLS DS-TE sub-pool
!
interface Tunnel1
!
!
ip address 10.20.2.2 255.255.255.252
! Same IP address as Red
!
ip address 10.20.2.2 255.255.255.252
! Same IP address as Blue
!
interface POS5/0
ip address 20.1.12.2 255.255.255.252
tag-switching ip
! Sub-pool defined on int
!
interface POS6/0
611

ip address 20.1.23.1 255.255.255.252
tag-switching ip
! No change to PE-P RSVP
!
router ospf 100
network 20.1.12.0 0.0.0.3 area 0
network 20.1.23.0 0.0.0.3 area 0
!
router bgp 100
no synchronization
no auto-summary
!
! Applies BGP PBR
neighbor 20.2.2.2 route-map TUNNEL-ASSIGNMENT in
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
ip classless
!
continues
612

!
!
!
access-list 2
10.1.1.0
0.0.0.255
1 permit
access-list
2 permit
10.1
!
match ip address 1
!
match ip address 2
!
match ip address 3
!

!
!
!
!
!
!
Identifies
Identifies
Identifies
Identifies
Identifies
Identifies
(Blue) Voice-VLAN
(Blue) Data-VLAN
(Blue) PE-CE link
(Red) Voice-VLAN
(Red) Data-VLAN
(Red) PE-CE Link



A P routers conguration for this MPLS DS-TE example is shown in Example 15-26 to
highlight the MPLS DS-TE congurations required in the core.
Example 15-26 P-Router MPLS DS-TE Example Conguration
!
hostname P-Router
!
!
ip cef
!
!
613
Example 15-26 P-Router MPLS DS-TE Example Conguration (Continued)

!
!
class CORE-REALTIME
priority percent 35
class class-default
fair-queue
!
!
interface Loopback0
ip address 20.3.3.3 255.255.255.255
!
interface POS5/0
ip address 20.1.13.2 255.255.255.252
tag-switching ip
!
interface POS6/0
ip address 20.1.23.2 255.255.255.252
tag-switching ip
!
router ospf 100
! MPLS TE RID
network 20.1.13.0 0.0.0.3 area 0
network 20.1.23.0 0.0.0.3 area 0
!

show mpls interface
614

show mpls trafc-eng tunnels
show mpls trafc-eng topology
Verication Command: show mpls trafc-eng topology

The show mpls trafc-eng topology command details (among other things) the amount
of allocated bandwidth for reservations on a per-priority basis. Example 15-27 shows the
details of the 54.25 Mbps (or 35 percent of the link) to be reserved in a subpool of priority
0 (best priority) on the PE-to-PE tunnel (Tunnel0). Furthermore, because the total reservable bandwidth on the PE-to-PE links is 77.5 Mbps, there remains 23.35 Mbps (77.5
54.25 Mbps) available to be reserved over the same link by global-pool trafc.
The command also shows that 77.5 Mbps of global-pool trafc has been allocated over the
PE-to-P tunnel (Tunnel1).
Example 15-27 Verication of Subpool and Global-Pool RSVP Reservations for DS-TE
PE1#show mpls traffic-eng topology 20.1.1.1
IGP Id: 20.1.1.1, MPLS TE Id:20.1.1.1 Router Node id 1
link[0 ]:Nbr IGP Id: 20.2.2.2, nbr_node_id:2, gen:30
frag_id 0, Intf Address:20.1.12.1, Nbr Intf Address:20.1.12.2
TE metric:1, IGP metric:1, attribute_flags:0x0
physical_bw: 155000 (kbps), max_reservable_bw_global: 77500 (kbps)
max_reservable_bw_sub: 54250 (kbps)
Global Pool
Sub Pool
Total Allocated
Reservable
Reservable
BW (kbps)
BW (kbps)
BW (kbps)
---------------------------------bw[0]:
54250
23250
0
bw[1]:
0
23250
0
bw[2]:
0
23250
0
bw[3]:
0
23250
0
bw[4]:
0
23250
0
bw[5]:
0
23250
0
bw[6]:
0
23250
0
bw[7]:
0
23250
0
link[1 ]:Nbr IGP Id: 20.3.3.3, nbr_node_id:3, gen:32
frag_id 1, Intf Address:20.1.13.1, Nbr Intf Address:20.1.13.2
TE metric:1, IGP metric:1, attribute_flags:0x0
physical_bw: 155000 (kbps), max_reservable_bw_global: 77500 (kbps)
max_reservable_bw_sub: 0 (kbps)
615
Example 15-27 Verication of Subpool and Global-Pool RSVP Reservations for DS-TE (Continued)
bw[0]:
bw[1]:
bw[2]:
bw[3]:
bw[4]:
bw[5]:
bw[6]:
bw[7]:
Total Allocated
BW (kbps)
--------------0
0
0
0
0
0
0
77500
Global Pool
Reservable
BW (kbps)
----------77500
77500
77500
77500
77500
77500
77500
0
Sub Pool
Reservable
BW (kbps)
---------0
0
0
0
0
0
0
0
PE1#
Verication Command: show ip bgp vpnv4 all

The show ip bgp vpnv4 all command displays the BGP next-hop (and other) attributes of
all prexes from all VRFs. In Example 15-28, the next-hop attribute of the prex representing Customer Blues adjacent CEs voice VLAN (10.1.102.0/24) is shown to be set to the
ctitious address of 16.16.16.16, which is resolved recursively to point to Tunnel0 (the premium DS-TE tunnel for voice). Although Customer Red has the same IP address prex, its
next hop is set to the ctitious address of 17.17.17.17, which is resolved recursively to point
to Tunnel1 (the default tunnel).
Example 15-28 Verication of BGP Next-Hop Policies
PE1#show ip bgp vpnv4 all
BGP table version is 78, local router ID is 20.1.1.1
Status codes: s suppressed, d damped, h history, * valid, > best, i - internal,
r RIB-failure, S Stale
Origin codes: i - IGP, e - EGP, ? - incomplete
Network
Next Hop
Metric LocPrf Weight Path
Route Distinguisher: 100:1 (default for vrf BLUE)
*> 10.1.1.0/24
10.20.1.1
0
0 10 ?
*> 10.1.101.0/24
10.20.1.1
0
0 10 ?
*>i10.2.2.0/24
17.17.17.17
0
100
0 10 ?
*>i10.2.102.0/24
16.16.16.16
0
100
0 10 ?
* 10.20.1.0/30
10.20.1.1
0
0 10 ?
*>
0.0.0.0
0
32768 ?
*>i10.20.2.0/30
17.17.17.17
0
100
0 ?
Route Distinguisher: 150:1 (default for vrf RED)
*> 10.1.1.0/24
10.20.1.1
0
0 15 ?
*> 10.1.101.0/24
10.20.1.1
0
0 15 ?
*>i10.2.2.0/24
17.17.17.17
0
100
0 15 ?
*>i10.2.102.0/24
17.17.17.17
0
100
0 15 ?
* 10.20.1.0/30
10.20.1.1
0
0 15 ?
*>
0.0.0.0
0
32768 ?
*>i10.20.2.0/30
17.17.17.17
0
100
0 ?
PE1#
616

Continuing the example from the previous design chapters, the ctitious company ABC,
Inc., has been growing and expanding, both geographically and technologically. It has multiple data centers in geographically diverse regions to which its eld needs to connect efciently. Additionally, to increase collaboration and simultaneously reduce travel expenses,
ABC, Inc., plans to roll out any-to-any videoconferencing. For these business reasons, ABC,
Inc., has decided to migrate from its private WAN to an MPLS VPN, managed by service
provider XYZ (SP XYZ).
To minimize the costs of migration, SP XYZ supports both Frame Relay and ATM Layer 2
access (including ATM IMA, which has been ABC, Inc.s, primary choice for branch WAN
media).
Furthermore, SP XYZ is a leader in MPLS VPN services and supports a ve-class provideredge model. Real-time service can be purchased in 5 percent increments, as can the
amounts of the three other levels of preferred service (Critical Data, Video, and Bulk Data).
ABC Inc. wants its WAN migration to MPLS VPN to be as transparent to end users as
possible, so it agrees to purchase these services in amounts that closely match the current
QoS Baseline WAN edge model, without causing trafc class-mixing issues.
Additionally, ABC, Inc., monitors network utilization (particularly videoconferencing
trafc) and performs trafc accounting and department bill-back based on the DSCP
markings of trafc ows. ABC, Inc., views it as essential that the SP not re-mark any trafc
at Layer 3 as it traverses the MPLS VPN, but rather preserve the DSCP markings intact.
Again, SP XYZ can accommodate ABC, Inc., because it deploys the popular Short Pipe
Mode of MPLS DiffServ tunneling.
SP XYZ also offers the option of premium service for voice trafc to geographically
adjacent sites (through MPLS DS-TE). Because ABC, Inc., is a heavy IP telephony user, it
elects to purchase this premium service for voice trafc (thus, from SP XYZs perspective,
ABC, Inc., is considered a BLUE class customer).
Figure 15-22 illustrates the combined enterprise and service-provider nodes where QoS
policies are required for this complex MPLS VPN scenario.
In Example 15-29, it is assumed that trafc has been marked correctly on campus/branch
switches before it arrives at the CE LAN edges. Where such an assumption is invalid,
ingress LAN edge marking policies, discussed in Chapter 14, Branch Router QoS
Design, can be applied to the CE LAN edges. Additionally, it has been assumed that
there are no unidirectional applications in this example.
Queuing and marking policies for a ve-class provider-edge model have been applied on
CE edges.
617
Figure 15-22 Case Study: MPLS VPN QoS Design Example Details
MPLS DS-TE policies:
Customer Blue Voice traffic is sent over direct link
to adjacent PEs (not through core).
Five-Class CE-PE
Mapping Policies
ABCs CE1 Router
(SP customer class: Blue)
10.20.1.0/30
10.1.1.0/24
10.1.101.0/24
PE1
PE2
Tunnel 0
ABCs CE2 Router

(SP customer class: Blue)
10.20.2.0/30
10.2.2.0/24
10.2.102.0/24
20.1.12.0/30
10.20.1.0/30
DSCP-to-CoS
Remapping Policy
Core DiffServ Policies:

Three-Class Per-Link
Scheduling Policies
30
0/
3.
.1
.1
20
Red CE1
Router
Lo: 20.1.1.1/32
Lo: 20.2.2.2
20
.1
.2
3.
0/
30
10.1.1.0/24
10.1.101.0/24
Tunnel 1
Short-Pipe Mode
ingress policies:
MPLS EXP values
are explicitly set
by service provider.
10.20.2.0/30
Red CE2 Router
10.2.2.0/24
10.2.102.0/24
Short-Pipe Mode egress policies:

Five-Class PE-CE scheduling/
dropping policies are based on
customers DiffServ markings.
Lo: 20.3.3.3
P Router
MPLS VPN Core
MPLS DS-TE policies:

Customer Blue data traffic and all Customer
Red traffic (both Voice and Data) are always sent
through the core.
On ingress, SP XYZ applies a ve-class short pipe MPLS DiffServ tunneling mode policer
to identify (through MPLS EXP values) trafc that is in contract or out-of-contract.
DiffServ policies are applied throughout the MPLS VPN core, and MPLS DS-TE also is
provisioned for voice trafc to geographically adjacent CEs. On egress, SP XYZ applies a
ve-class provider-edge model, which is based on the customers DiffServ markings. In this
example, company ABC, Inc., ts service provider XYZs customer Blue prole.
The conguration for this example spans six routers: Blue-CE1, Blue-CE2, Red-CE1,
Red-CE2, PE1, PE2, and P router. However, because CE congurations are virtually identical, only one is presented here (Blue-CE1see Example 15-29), along with the
congurations for PE1 (see Example 15-30), PE2 (see Example 15-31), and the P router
(see Example 15-32).
Example 15-29 Blue-CE1 Case Study MPLS VPN QoS Design Example
!
hostname CE1-BLUE
!
ip cef
!
match ip dscp cs6
! IP CEF is required for Packet Marking
continues
618
Example 15-29 Blue-CE1 Case Study MPLS VPN QoS Design Example (Continued)
match ip dscp ef
match ip dscp af41
match ip dscp cs4
match ip dscp 25
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp af11
match ip dscp cs2
match ip dscp cs1
!
!
policy-map CE-FIVE-CLASS-SP-MODEL
class ROUTING
class VOICE
priority percent 15
set ip dscp cs5
set ip dscp af21
priority percent 2
set ip dscp cs5
random-detect
set ip dscp af31
bandwidth percent 5
random-detect
set ip dscp cs3
class BULK-DATA
bandwidth percent 5 ! Bulk Data is assigned to Bulk SP class
random-detect
class SCAVENGER
bandwidth percent 1
set ip dscp 0
619
class class-default
random-detect
!
!
policy-map CE-LAN-EDGE-OUT
class class-default
set cos dscp
! Enables default DSCP-to-CoS Mapping

!
!
description TO CAT3500 BRANCH ACCESS-SWITCH
no ip address
!
description DLVAN SUBNET 10.1.1.0
ip address 10.1.1.1 255.255.255.0
service-policy output CE-LAN-EDGE-OUT
! Restores CoS for Data VLAN
!
!
ip address 10.1.101.1 255.255.255.0
service-policy output CE-LAN-EDGE-OUT
!
!
interface ATM1/0
no ip address
ima-group 1
!
interface ATM1/1
no ip address
ima-group 1
!
!
interface ATM1/IMA1
no ip address
!
description Dual-T1 ATM IMA Link to PE1
ip address 10.20.1.1 255.255.255.252
pvc 0/120
vbr-nrt 3072 3072
continues
620
service-policy output CE-FIVE-CLASS-SP-MODEL
!
! Overrides 75% BW limit

! Applies 5-Class CE-PE Model
!
router bgp 10
no synchronization
no auto-summary
!
!
The conguration for the rst PE router for this MPLS VPN QoS design case study
example is shown in Example 15-30.
Example 15-30 PE1 Case Study MPLS VPN QoS Design Example
!
hostname PE1
!
!
ip vrf BLUE
rd 100:1
!
ip vrf RED
rd 150:1
!
ip cef
!
!
!
match ip dscp ef
match ip dscp cs5
match ip dscp cs6
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp cs2
match ip dscp af11
match ip dscp cs1
! BLUE MPLS VPN Definition
! RED MPLS VPN Definition
! Enables MPLS TE globally
621
Example 15-30 PE1 Case Study MPLS VPN QoS Design Example (Continued)
match mpls experimental topmost 5 ! Identifies in-contract Realtime
match mpls experimental topmost 3 ! Identifies in-contract Critical-Data
match mpls experimental topmost 7 ! Identifies out-of-contract Critical Data
match mpls experimental topmost 2 ! Identifies in-contract Video
match mpls experimental topmost 1 ! Identifies in-contract Bulk
match mpls experimental topmost 6 ! Identifies out-of-contract Bulk
!
!
class
REALTIME
exceed-action
set-mpls-exp-topmost-transmit 7
police cir 1050000
exceed-action drop
class CRITICAL-DATA
police cir 600000
class VIDEO
police cir 450000
exceed-action drop
class BULK-DATA
police cir 150000
class class-default
police cir 750000
!
!
class REALTIME
priority percent 35
class CRITICAL-DATA
class VIDEO
class BULK-DATA
bandwidth percent 5
class class-default
random-detect
!
continues
622
!
class CORE-REALTIME
priority percent 35
class class-default
fair-queue
! CORE-BEST-EFFORT gets FQ
!
!
interface Loopback0
ip address 20.1.1.1 255.255.255.255
!
interface Tunnel0
! Enables MPLS TE on tunnel
! Best priority
tunnel mpls traffic-eng bandwidth sub-pool 54250 ! Assigns sub-pool
!
interface Tunnel1
! Enables MPLS TE
! Worst priority
! Assigns global pool
!
!
interface ATM2/0
no ip address
ima-group 1
!
interface ATM2/1
no ip address
ima-group 1
!
interface ATM2/ima1
no ip address
!
interface ATM2/ima1.20 point-to-point
description Dual-T1 ATM IMA Link to Blue CE1
ip address 10.20.1.2 255.255.255.252
pvc 0/120
vbr-nrt 3072 3072
! Overrides 75% BW
623
service-policy output PE-FIVE-CLASS-SP-MODEL
!
! Short Pipe Marking

! Egress policy to CE
!
interface ATM2/2
no ip address
ima-group 2
!
interface ATM2/ima2
no ip address
!
description Dual-T1 ATM IMA Link to Red CE1
ip address 10.20.1.2 255.255.255.252
pvc 0/220
vbr-nrt 3072 3072
! Overrides 75% BW
!
!
interface ATM2/3
no ip address
ima-group 2
!
!
interface POS5/0
ip address 20.1.12.1 255.255.255.252
! Applies Core DS policies
! Enables MPLS TE on int
tag-switching ip
! Assigns sub-pool BW
!
interface POS6/0
ip address 20.1.13.1 255.255.255.252
tag-switching ip
! Assigns global-pool BW
!
router ospf 100
! MPLS TE RID
continues
624
network 20.1.12.0 0.0.0.3 area 0
network 20.1.13.0 0.0.0.3 area 0
!
router bgp 100
no synchronization
no auto-summary
!
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
ip classless
!
!
! Defines explicit path for Tu0
!
!
625
!
match ip address 1
!
match ip address 2
!
match ip address 3
!
!
!
!
!
!
!
!
Identifies
Identifies
Identifies
Identifies
Identifies
Identifies
(Blue) Voice-VLAN
(Blue) Data-VLAN
(Blue) PE-CE link
(Red) Voice-VLAN
(Red) Data-VLAN
(Red) PE-CE Link



Example 15-31 shows the conguration for the second PE router for this MPLS VPN QoS
design case study example.
Example 15-31 PE2 Case Study MPLS VPN QoS Design Example
!
hostname PE2
!
!
ip vrf BLUE
rd 100:1
!
ip vrf RED
rd 150:1
!
ip cef
!
!
! BLUE MPLS VPN Definition
! RED MPLS VPN Definition
! Enables MPLS TE globally
continues
626
!
match ip dscp ef
match ip dscp cs5
match ip dscp cs6
match ip dscp af31
match ip dscp cs3
match ip dscp af21
match ip dscp cs2
match ip dscp af11
match ip dscp cs1
match mpls experimental topmost 5 ! Identifies in-contract Realtime
match mpls experimental topmost 3 ! Identifies in-contract Critical-Data
match mpls experimental topmost 7 ! Identifies out-of-contract Critical Data
match mpls experimental topmost 2 ! Identifies in-contract Video
match mpls experimental topmost 1 ! Identifies in-contract Bulk
match mpls experimental topmost 6 ! Identifies out-of-contract Bulk
!
!
class REALTIME
police cir 1050000
exceed-action drop
class CRITICAL-DATA
police cir 600000
class VIDEO
police cir 450000
exceed-action drop
class BULK-DATA
police cir 150000
class class-default
police cir 750000
!
!
class REALTIME
priority percent 35
class CRITICAL-DATA
627
class VIDEO
class BULK-DATA
bandwidth percent 5
class class-default
random-detect

!
!
class CORE-REALTIME
priority percent 35
class class-default
fair-queue
! CORE-BEST-EFFORT gets WFQ
!
!
interface Loopback0
ip address 20.2.2.2 255.255.255.255
!
interface Tunnel0
! Enables MPLS TE on tunnel
! Best priority
! Assigns sub-pool
!
interface Tunnel1
! Enables MPLS TE
! Worst priority
! Assigns global pool
!
!
interface ATM2/0
no ip address
ima-group 1
!
interface ATM2/1
no ip address
ima-group 1
continues
628
!
interface ATM2/ima1
no ip address
!
description Dual-T1 ATM IMA Link to Blue CE2
ip address 10.20.2.2 255.255.255.252
pvc 0/120
vbr-nrt 3072 3072
! Overrides 75% BW
!
!
interface ATM2/2
no ip address
ima-group 2
!
interface ATM2/ima2
no ip address
!
description Dual-T1 ATM IMA Link to Red CE2
ip address 10.20.2.2 255.255.255.252
pvc 0/220
vbr-nrt 3072 3072
! Overrides 75% BW
!
!
interface POS5/0
ip address 20.1.12.2 255.255.255.252
tag-switching ip
! Assigns sub-pool BW
!
interface POS6/0
ip address 20.1.23.1 255.255.255.252
tag-switching ip
629
!
router ospf 100
!
!
network 20.1.12.0 0.0.0.3 area 0
network 20.1.23.0 0.0.0.3 area 0
!
router bgp 100
no synchronization
no auto-summary
!
neighbor 20.1.1.1 route-map TUNNEL-ASSIGNMENT
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
no auto-summary
no synchronization
exit-address-family
!
ip classless
ip route 18.18.18.18 255.255.255.255 Tunnel0 !
ip route 19.19.19.19 255.255.255.255 Tunnel1 !
!
!
!
!
MPLS TE RID
Enables OSPF area 0 for MPLS TE
in
! Applies BGP PBR
Static route for Tunnel 0

Static route for Tunnel 1
Identifies Blue VPN by RT
Identifies Red VPN by RT
continues
630
!
!
!
match ip address 1
!
match ip address 2
!
match ip address 3
!
!
!
!
!
!
!
!
Identifies
Identifies
Identifies
Identifies
Identifies
Identifies
(Blue) Voice-VLAN
(Blue) Data-VLAN
(Blue) PE-CE link
(Red) Voice-VLAN
(Red) Data-VLAN
(Red) PE-CE Link



The conguration for the P router for this MPLS VPN QoS design case-study example is
Example 15-32 P-Router Case Study MPLS VPN QoS Design Example
!
hostname P-Router
!
!
ip cef
!
!
match mpls experimental topmost 5 !
Identifies in-contract Realtime

Identifies in-contract Critical-Data
Identifies out-of-contract Critical Data
Identifies in-contract Video
631
Example 15-32 P-Router Case Study MPLS VPN QoS Design Example (Continued)
! Identifies in-contract Bulk

! Identifies out-of-contract Bulk
!
!
class CORE-REALTIME
priority percent 35
class class-default
fair-queue
! CORE-BEST-EFFORT gets WFQ
!
!
interface Loopback0
ip address 20.3.3.3 255.255.255.255
!
!
interface POS5/0
ip address 20.1.13.2 255.255.255.252
tag-switching ip
!
interface POS6/0
ip address 20.1.23.2 255.255.255.252
tag-switching ip
!
router ospf 100
! MPLS TE RID
network 20.1.13.0 0.0.0.3 area 0
network 20.1.23.0 0.0.0.3 area 0
!
!

show mpls interface
632

show mpls trafc-eng tunnels
show mpls trafc-eng topology
Summary
MPLS VPNs are rapidly gaining popularity as private WAN alternatives. This chapter
presented QoS design principles and designs to achieve end-to-end service levels over
MPLS VPNs.. The foremost design principle is that enterprise subscribers and service
providers have to cooperatively deploy QoS over MPLS VPNs in a consistent and
complementary manner.
Enterprise (customer) considerations, such as class-collapsing guidelines and trafcmixing principles, were overviewed along with re-marking examples.
Service-provider edge QoS policies were presented for three-, four-, and ve-class edge
models. Additionally, details on how RFC 3270 tunneling modes (Uniform, Short Pipe, and
Pipe) can be implemented within Cisco IOS Software were provided.
Service provider core QoS options, such as aggregate bandwidth overprovisioning or
deploying DiffServ in the backbone, were considered along with platform-specic
considerations, such as MDRR on the 12000 GSR.
MPLS trafc engineering was introduced in brief, and two examples of MPLS TE for QoS
were presented (MPLS per-VPN TE and MPLS DS-TE).
The chapter concluded with a case-study continuation from previous chapters, showing
how an enterprise that is compliant with the QoS Baseline can migrate its private WAN to
an MPLS VPN. The service providers QoS designs also were detailed within the case
study.
Further Reading
Standards:
RFC 2547, BGP/MPLS VPNs: http://www.ietf.org/rfc/rfc2547.txt.
RFC 2917, A Core MPLS IP VPN Architecture: http://www.ietf.org/rfc/rfc2917.txt.
RFC 2597, Assured Forwarding PHB Group: http://www.ietf.org/rfc/rfc2597.txt.

RFC 2702, Requirements for Trafc Engineering over MPLS: http://www.ietf.org/
rfc/rfc2702.txt.
Further Reading
RFC 3270, Multiprotocol Label Switching (MPLS) Support of Differentiated

Services: http://www.ietf.org/rfc/rfc3270.txt.
RFC 3564, Requirements for Support of Differentiated Services-Aware MPLS

Trafc Engineering: http://www.ietf.org/rfc/rfc3564.txt.
633
Books:
Alwayn, Vivek. Advanced MPLS Design and Implementation. Indianapolis: Cisco

Press, 2001.
Pepelnjak, Ivan, and Jim Guichard. MPLS and VPN Architectures. Cisco Press, 2002.
Osborne, Eric, and Ajay Simha. Trafc Engineering with MPLS. Cisco Press, 2003.
Pepelnjak, Ivan, Jim Guichard, and Jeff Apcar. MPLS and VPN Architectures. Cisco
Press, 2003.
Cisco MPLS features:
Conguring MPLS and MPLS trafc engineering (Cisco IOS Release 12.2): http://
www.cisco.com/univercd/cc/td/doc/product/software/ios122/122cgcr/fswtch_c/
swprt3/xcftagc.htm.
MPLS VPNS (Cisco IOS Release 12.2.13T): http://www.cisco.com/univercd/cc/td/

doc/product/software/ios122/122newft/122t/122t13/ftvpn13.htm.
MPLS DiffServ tunneling modes (Cisco IOS Release 12.2.13T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122newft/122t/122t13/ftdtmode.htm.
MPLS DiffServ-Aware Trafc Engineering (DS-TE) (Cisco IOS Release 12.2.4T):

122t4/ft_ds_te.htm.
MPLS Cisco IOS documentation main link (Cisco IOS Release 12.3):
swit_vcg.htm#999526.
IPSec VPNs are the most widely deployed VPNs and are found in three main contexts:
Site-to-site IPSec VPNs

Teleworker IPSec VPNs
Remote-access client (mobility) IPSec VPNs
QoS considerations for site-to-site and teleworker IPSec VPNs are examined in this design
chapter (as QoS is rarelyif everdeployed in remote-access client IPSec VPN
scenarios). These considerations include the following:
IPSec modes of operation

Bandwidth and delay increases because of encryption
IPSec and cRTP incompatibility
IP ToS byte preservation through IPSec encryption
QoS and Anti-Replay interaction implications
Following a discussion of these considerations, design recommendations for site-to-site

and teleworker (DSL and cable) solutions are presented in detail.
CHAPTER
16

Whereas MPLS technologies provide VPN services, such as network segregation and
privacy, by maintaining independent virtual router forwarding tables, IPSec achieves such
VPN services through encryption.
As dened in RFCs 2401 through 2412, IPSec protocols provide mechanisms to enable
remote sites to cost effectively connect to enterprise intranets or extranets using the Internet
(or a service providers shared IP networks). Because of IPSec protocol encryption, such
VPNs can provide the same management and security policies as private networks.
IPSec VPN services are built by overlaying a point-to-point mesh over the Internet using
Layer 3encrypted tunnels. This architecture requires security appliances, such as (hardware or software) rewalls or routers that support IPSec tunnels, to be installed at both ends
of each tunnel. Encryption/decryption is performed at these tunnel endpoints, and the protected trafc is carried across the shared network.
Three main design contexts for IPSec VPNs exist, as shown in Figure 16-1:
Site-to-site VPNsTunnels are maintained by hardware devices to connect multiple

users at remote branches to (one or more) central sites.
Teleworker VPNsTunnels are maintained by hardware devices to connect

(typically) a single user at his or her residence to a central site.
Remote-access clientsTunnels are established by software to connect mobile users

at airports, hotels, or similar places to a central site using WLAN hotspots, LAN ports,
or modems.
636
Chapter 16: IPSec VPN QoS Design
Figure 16-1 IPSec VPN Design Contexts
Site-to-Site
IPSec VPN
Central Site
Service Provider/
Internet
WAN
Aggregators
IPSec VPN
Headers
Teleworker
IPSec VPN
IPSec VPN
Tunnels
Remote Access
IPSec VPN
(Software Client)
Enabling converged services, such as voice and video, on an IPSec VPN has been dubbed
V3PN. V3PN is essentially the overlaying of QoS technologies over IPSec VPNs to provide
the required service levels to voice and video applications. As such, V3PN solutions relate
to only two of the three IPSec VPN design contexts: site-to-site VPNs and telecommuter
VPNs. (Little, if any, QoS is available in remote-access client networks.)
This chapter discusses QoS design considerations and recommendations for both site-tosite and telecommuter V3PN solutions.
NOTE
It is beyond the scope of this chapter to detail IPSec encryption operation and conguration;
a working knowledge of IPSec is assumed.
Site-to-Site V3PN QoS Considerations
637

Attractive pricing is usually the driver behind deploying site-to-site IPSec VPNs as an
alternative to private WAN technologies. Many of the same considerations required by
private WANs need to be taken into account for IPSec VPN scenarios because they usually
are deployed over the same Layer 2 WAN access media.
IPSec VPNs also share some similar concerns with MPLS VPNs. For instance, the
enterprises end-to-end delay and jitter budgets depend signicantly on the service
providers SLAs. Therefore, enterprises deploying V3PN solutions are recommended
to utilize Cisco Powered Network IP Multiservice service providers, as discussed in
Chapter 15, MPLS VPN QoS Design.
However, IPSec VPNs present many unique considerations for QoS design, including the
following (each is discussed in detail throughout the rest of the chapter):
IPSec VPN modes of operation

Packet overhead increases because of encryption
cRTP and IPSec incompatibility
Prefragmentation
Bandwidth provisioning
Logical topologies
Delay budget increases because of encryption
ToS byte preservation
QoS Pre-Classify feature
Pre-encryption queuing
Anti-Replay implications
Control plane provisioning
IPSec VPN Modes of Operation

Three principal modes of IPSec VPN operation exist:
IPSec tunnel modeNo IP GRE tunnel

IPSec transport modeEncrypting an IP GRE tunnel
IPSec tunnel modeEncrypting an IP GRE tunnel
The advantages, disadvantages, features, and limitations of these options are discussed
next.
638
IPSec Tunnel Mode (No IP GRE Tunnel)

This option does not utilize an IP GRE tunnel. With this option, only IPSec unicast trafc
can be transported. (IP multicast trafc cannot be transported between IPSec peers without
conguring an IP GRE tunnel.)
This conguration might be sufcient to support application requirements; its advantage
lies in lower CPU overhead (primarily at the headend IPSec VPN router) compared with
alternative IPSec design options.
IPSec security associations (SAs) are created for each access list line matched. An access
list must be specied in the crypto map to designate packets that are to be encrypted. Such
an access list typically entails several lines to dene the application(s) to be encrypted by
the ve ACL tuples: source/destination IP address, protocol, and source/destination port
numbers. When not encrypting a GRE tunnel, it is possible to create a separate SA for each
application or access-list line match or to create an SA that carries all trafc that matches
an ACL range (which is recommended). Each SA has its own Encryption Security Protocol
(ESP) or Authentication Header (AH) sequence number.
Anti-Replay drops can be eliminated or minimized by constructing access lists that create
a separate security association for each class of trafc being inuenced by per-hop QoS
policies. (Anti-Replay is an IPSec standard feature that discards packets that fall outside a
receivers 64-byte sliding window because such packets are considered suspect or potentially compromisedit is discussed in greater detail later in this chapter.)
The Cisco IOS feature of prefragmentation for IPSec VPNs (also discussed later in this
chapter) is supported in IPSec tunnel mode (no IP GRE tunnel) as of Cisco IOS Release
12.2(12)T.
IPSec Transport Mode (Encrypting an IP GRE Tunnel)

IPSec transport mode (encrypting an IP GRE tunnel) is a commonly deployed option because
it provides all the advantages of using IP GRE, such as IP Multicast protocol support (and,
thus, also the support of routing protocols that utilize IP Multicast) and multiprotocol support. Furthermore, this option saves 20 bytes per packet over IPSec tunnel mode (encrypting
an IP GRE tunnel) because an additional IP header is not required. Figure 16-2 illustrates
IPSec transport mode versus tunnel mode when encryption is performed in an IP GRE
tunnel.
The IPSec peer IP addresses and the IP GRE peer address must match for transport mode
to be negotiated; if they do not match, tunnel mode is negotiated.
639
Figure 16-2 IPSec Transport Mode Versus Tunnel Mode for a G.729 VoIP Packet
IPSec ESP
Transport Mode
120 Bytes
IPSec ESP
Tunnel Mode
140 Bytes
IPSec
Hdr
ESP
Hdr
ESP
IV
GRE
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
20
20
12
20
2-257
12
IPSec
Hdr
ESP
Hdr
ESP
IV
GRE IP
Hdr
GRE
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
20
20
20
12
20
2-257
12
The Cisco IOS prefragmentation feature for IPSec VPNs (discussed later in the chapter) is
not supported for transport mode because the decrypting router cannot determine whether
the fragmentation was done before or after encryption (for example, by a downstream
router between the encrypting and decrypting routers).
Although IPSec transport mode saves a small to moderate amount of link bandwidth, it does
not provide any reduction in packets per second switched by the router. Therefore, because
the number of packets per second primarily affects CPU performance, no signicant CPU
performance gain is realized by using IPSec transport mode.
IPSec tunnel mode is the default conguration option. To congure transport mode, it must
be specied under the IPSec transform set, as shown in Example 16-1.
Example 16-1 Enabling IPSec Transport Mode
!
crypto ipsec transform-set ENTERPRISE esp-3des esp-sha-hmac
! Enables IPSec Transport mode
mode transport
!
IPSec Tunnel Mode (Encrypting an IP GRE Tunnel)

IPSec tunnel mode (encrypting an IP GRE tunnel) is the primarily recommended IPSec
VPN design option. Although it incurs the greatest header overhead of the three options, it
is capable of supporting IP Multicast (with the capability to run a dynamic routing protocol
within the IP GRE tunnel for failover to an alternative path), and it supports prefragmentation for IPSec VPNs.
When congured with a routing protocol running within an IP GRE tunnel, the routing
protocols Hello packets maintain the security associations between the branch and both
(assuming a redundant conguration) headend routers. There is no need to create a security
association with a backup headend peer if the primary peer fails.
640
NOTE
The design principles in this chapter were proven by scalability testing in the Cisco
Enterprise Solutions Engineering labs. These large-scale testing methods were designed to
test worst-case scenarios. From a design standpoint, these entailed enabling the following:
Strong Triple-Digital Encryption Standard (3DES) encryption for both Internet Key
Exchange (IKE) and IPSec
IP GRE with IPSec tunnel mode
Dife-Hellman Group 2 (1024 bit) for IKE
Secure Hash Algorithm (SHA) 160-bit RFC 2104 Keyed-Hashing for Message
Authentication (HMAC) with RFC 1321 Message Digest 5 (MD5)

(MD5)-HMAC (both hash algorithms truncated to 12 bytes in the ESP packet trailer)
Preshared keys
If an enterprise chooses to implement less stringent security parameters, to use IPSec

transport mode instead of tunnel mode, or to not implement IP GRE tunnels, the designs
continue to be applicable from functional and scalability standpoints.
Packet Overhead Increases

The addition of tunnel headers and encryption overhead increases the packet sizes of all
encrypted applications: voice, video, and data. This needs to be taken into account when
provisioning LLQ or CBWFQ bandwidth to a given class.
For example, consider voice. The two most widely deployed codecs for voice are G.711
and G.729. Each codec typically is deployed at 50 pps (generating 20-ms packetization
intervals).
The Layer 3 data rate for a G.711 call (at 50 pps) is 80 kbps. IP Generic Routing Encapsulation (GRE) tunnel overhead adds 24 bytes per packet. The IPSec Encapsulating Security
Payload (ESP) adds another 56 bytes. The combined additional overhead increases the rate
from 80 kbps (clear voice) to 112 kbps (IPSec ESP tunnel-mode encrypted voice).
The calculation is as follows:
200 bytes per packet (G.711 voice)

24 bytes per packet (IP GRE overhead)
56 bytes per packet (IPSec ESP overhead)
280 bytes per packet

8 bits per byte
2240 bits per packet

50 packets per second
112,000 bits per second
641
The additional overhead represents a 40 percent increase in the bandwidth required for an
encrypted G.711 call.
The 280-byte packets header, data, and trailer elds for an IPSec tunnel-mode ESP
encrypted G.711 call are shown in Figure 16-3.
Figure 16-3 Anatomy of an IPSec-Encrypted G.711 Packet
G.711
200 Bytes
IP GRE
224 Bytes
GRE IP
GRE
Hdr
20
IPSec ESP
GRE
Tunnel Mode Hdr
Hdr
Hdr IV
280 Bytes
20
20
IP
Hdr
UDP
RTP
Voice
20
12
160
IP
Hdr
UDP
RTP
Voice
20
12
160
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
20
12
160
2257
12
Encrypted
Authenticated
The Layer 3 data rate for a G.729 call (at 50 pps) is 24 kbps. IP GRE tunnel overhead adds
24 bytes per packet. IPSec ESP adds another 52 bytes. The combined additional overhead
increases the rate from 24 kbps (clear voice) to just less than 56 kbps (IPSec ESP tunnelmode encrypted voice).
The calculation is as follows:

136 bytes per packet

8 bits per byte

The additional overhead represents a 227 percent increase in the bandwidth required for an
encrypted G.729 call.
642
The 136-byte packets header, data, and trailer elds for an IPSec tunnel-mode ESP
encrypted G.729 call are shown in Figure 16-4.
Figure 16-4 Anatomy of an IPSec-Encrypted G.729 Packet
IP
Hdr
UDP
RTP
Voice
20
12
20
IP
Hdr
UDP
RTP
Voice
20
12
20
GRE
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
20
12
20
2257
12
G.729
60 Bytes
IP GRE
84 Bytes
GRE IP
GRE
Hdr
20
IPSec ESP
Tunnel Mode
136 Bytes

Hdr
Hdr
Hdr IV
20
20
Encrypted
Authenticated
It is important to note that these bandwidth allocations are Layer 3 bandwidth requirements
and do not include Layer 2 overhead (which is media dependent). Therefore, Layer 2
overhead needs to be added on top of the Layer 3 requirements in provisioning LLQ and
CBWFQ bandwidth. This is illustrated in Figure 16-5, where Ethernet overhead (Ethernet
plus 802.1Q trunking) and Frame Relay overhead are added and removed from the packet
in transit.
Key Layer 2 overhead values are reiterated in Table 16-1.
Table 16-1
Layer 2 Encapsulation Overhead

L2 Encapsulation
Overhead
Ethernet
14 bytes (+ 4 for 802.1Q)
Frame Relay
4 bytes (+ 4 for FRF.12)
MLP
10 bytes (+ 3 for MLP LFI)
ATM
5 bytes per 53-byte cell + cell padding (variable)
643
Figure 16-5 Packet Size Changes of a G.729 IPSec-Encrypted Packet
IPSec VPN
(Frame Relay
Access)
140
Payload
(20)
Layer 2
Size
FR (4)
60
20
78
78
Packet
Size (Bytes)
Ethernet (14)
IPSec (52)
Ethernet (14)
60
802.1Q (4)
GRE (24)
802.1Q (4)
IP (20)
IP (20)
IP (20)
IP (20)
IP (20)
UDP (8)
UDP (8)
UDP (8)
UDP (8)
UDP (8)
RTP (12)
RTP (12)
RTP (12)
RTP (12)
RTP (12)
Payload
(20)
Payload
(20)
Payload
(20)
Payload
(20)
Payload
(20)
20
Payload
(20)
Therefore, the calculation, inclusive of Layer 2 overhead, is as follows. This example

assumes that a G.729 call will be encrypted over a slow speed ( 768-kbps Frame Relay
link), which requires FRF.12 fragmentation and interleaving.

4 bytes per packet (FR overhead)
4 bytes per packet (FRF.12 overhead)
44 bytes per packet

8 bits per byte

57,600 bits per second (rounded up to 58 kbps)
In summary, it is important always to include Layer 2 overhead in accurate bandwidth

provisioning for IPSec-encrypted applications.
cRTP and IPSec Incompatibility

The signicant increases in bandwidth required by IPSec encryption lead many
administrators to consider the use of IP RTP header compression (cRTP) to offset
these increases.
644
However, one of the caveats of encryption is that key portions of the original IP packet that
could be referenced for QoS (and other) purposes are no longer readable. Such is the case
with cRTP.
cRTP and IPSec are inherently incompatible standards. The original IP/UDP/RTP header
already is encrypted by IPSec by the time the RTP compressor is called upon to perform
the compression. Therefore, because cRTP cannot associate the encrypted IP/UDP/RTP
packet with a known media stream, compression cannot occur and cRTP bandwidth savings
cannot be realized. The encrypted IP/UDP/RTP packet simply bypasses the compression
process and continues (uncompressed) to the transmit queue.
This is illustrated in Figure 16-6.
Figure 16-6 IPSec and cRTP Incompatibility
Identify RTP
Traffic
RTP
Traffic
!@#!
IPSec
Crypto
Engine
Compress RTP
Traffic
RTP
Compressor
Huh?
Transmit
Queue
!@#!
!@#!
$#@
!@#!
!@#!
Configured
Queuing
!@#!
Non-RTP
Traffic
The RTP compression engine cannot identify the stream as RTP,

(since it has been encrypted) and, thus, cannot compress the stream.
It is important to recognize that cRTP functions on a hop-by-hop basis, whereas IPSec can
span multiple intermediate (Layer 3) hops between IPSec endpoints. This distinction
further exacerbates incompatibility between the features.
Although developments are under way to address these incompatibilities, at the time of this
writing, cRTP cannot be utilized to achieve bandwidth savings in an IPSec VPN
environment.
Prefragmentation
A problem arises when a packet is nearly the size of the maximum transmission unit (MTU)
of the outbound link of the encrypting router and then is encapsulated with IPSec headers.
645
The resulting packet is likely to exceed the MTU of the outbound link. This causes packet
fragmentation after encryption, which makes the decrypting router reassemble in the
process path.
Cisco IOS Release 12.2(13)T introduced a new feature: prefragmentation for IPSec VPNs.
Prefragmentation increases the decrypting routers performance by enabling it to operate in
the high-performance CEF path instead of the process path.
This feature enables an encrypting router to predetermine the encapsulated packet size from
information available in transform sets, which are congured as part of the IPSec security
association. If it is predetermined that the packet will exceed the MTU of the output
interface, the packet is fragmented before encryption. This function avoids process-level
reassembly before decryption and helps improve decryption performance and overall IPSec
trafc throughput.
Prefragmentation for IPSec VPNs is enabled globally by default for Cisco VPN routers
running Cisco IOS Release 12.2(13)T or higher.
Chapter 2, QoS Design Overview, presented the 33 Percent LLQ Rule, along with the
design rationale behind the recommendation. Furthermore, the rule was expressed as
a conservative design recommendation that might not be valid under all constraints.
Provisioning for VoIP over IPSec on slow links sometimes poses constraints that might
preclude applying the 33 Percent LLQ Rule.
As shown in Table 16-2, the percentage of LLQ required on 64-, 128-, and 256-kbps links
for a single encrypted G.729 call exceeds the recommended 33 percent LLQ limit. Enterprises considering such deployments must recognize the impact on data trafc traversing
such links when encrypted voice calls were madespecically, data applications would
slow down signicantly. If that is considered an acceptable trade-off, not much can be said
or done. Otherwise, it is recommended to increase bandwidth to the point that encrypted
VoIP calls can be made and still fall within the 33 percent bandwidth recommendation for
priority queuing.
When planning the bandwidth required for a branch ofce, consider the number of concurrent calls traversing the IPSec VPN that the branch is expected to make during peak call
periods. This varies based on the job function of the employees located at a branch. For
example, an ofce of software engineers would be expected to make fewer calls than an
ofce of telemarketers. A typical active call ratio may be one active call for every six people
(1:6), but this could range from 1:4 or 1:10, depending on the job function of the employees.
Given the 512-kbps link from Table 16-2 as an example, with a target of 3 G.729 calls, this
link theoretically could support a branch ofce of between 12 and 30 people. As with all
other topologies, call admission control must be administered properly to correspond to the
QoS policies deployed within the network infrastructure.
646
Table 16-2
NOTE
G.729 Calls by Link Speeds (FRF.12 Is Enabled on Link Speeds 768 kbps Only)
Line Rate (kbps)
Maximum Number
of G.729 Calls
LLQ Bandwidth
(kbps)
LLQ Bandwidth
(Percentage)
64 (FRF.12)
1 (58 kbps)
58
91%
128 (FRF.12)
1 (58 kbps)
58
46%
256 (FRF.12)
2 (58 kbps)
116
46%
512 (FRF.12)
3 (58 kbps)
174
34%
768 (FRF.12)
4 (58 kbps)
232
31%
1024
6 (56 kbps)
336
33%
1536
9 (56 kbps)
504
33%
2048
12 (56 kbps)
672
33%
Although VoIP has been discussed as a primary example of bandwidth provisioning

considerations when deploying QoS over IPSec VPNs, it is important to recognize that
VoIP is not the only application that might require such considerations; this exercise needs
to be performed for any application that is being encrypted.
Unlike VoIP, however, other applicationssuch as video and datahave varying packet
rates and sizes. Therefore, crisp provisioning formulas might not apply. Moderate trafc
analysis and best guesses, along with trial-and-error tuning, are usually the only options for
factoring bandwidth provisioning increases for non-VoIP applications.
Logical Topologies
Similar to private WANs, but unlike fully meshed MPLS VPNs, IPSec VPNs typically are
deployed in a (logical) hub-and-spoke topology.
The QoS implications of hub-and-spoke topologies include that access rates to remote
sites need to be constrained and shaped at the hub, to avoid delays and drops within the
providers cloud. Shaping at the IPSec VPN hub is done in a manner similar to that of
private WAN NBMA media, such as Frame Relay or ATM. Refer to the Headend VPN
Edge QoS Options for Site-to-Site V3PNs section, later in this chapter, and also to
Chapter 13, WAN Aggregator QoS Design, for additional details.
IPSec VPNs are not limited to hub-and-spoke designs. They also can be deployed in partialmesh or even fully meshed topologies. In such cases, shaping is recommended on any links
where speed mismatches occur (similar to private WAN scenarios).
Another alternative is to deploy IPSec VPNs via Dynamic Multipoint Virtual Private Networks (DMVPN), which establish site-to-site IPSec tunnels as needed and tear them down
647
when they no longer are required. As with the previously discussed logical topologies,
shaping is required on DMVPN NBMA links with speed mismatches. Specically, shapers
are required to be created dynamically and applied to logical DMVPN tunnels to offset any
speed mismatches attributed to physical NMBA links. However, as of the time of this writing, no shaping or queuing solution exists to guarantee QoS SLAs over DMVPN topologies
(although Cisco IOS solutions currently are being evaluated and are in development).
Delay Budget Increases

As previously discussed, the delay budget for a typical IP Telephony implementation
includes xed and variable components. The ITU G.114 specications target value for
one-way delay is 150 ms. In an IPSec VPN deployment, however, two additional delay
components must be factored into the overall delay budget:
Encryption delay at the origination point of the IPSec VPN tunnel

Decryption delay at the termination point of the IPSec VPN tunnel
Performance and scalability testing results suggest that, in most cases, the additional delay
caused by encryption and decryption is approximately 4 to 10 ms (combined). These
incremental delays are shown in Figure 16-7.
Figure 16-7 IPSec Encryption/Decryption Incremental Delays
Service
Provider
Si
Si
Campus
CODEC
10-50 ms
(Depends on
Sample Size)
Branch
O
Offic
Office
Queuing
Variable
(Can Be Reduced
Using Hardware
PQs)
Encrypt
Minimal
2-5 ms
Queuing +
Serialization
Variable
(Can Be Reduced
Using LLQ+LFI)
Latency Target
Propagation and
Network
6.3 s / Km +
Network
Delay
(Variable)
Decrypt
Jitter Buffer
Minimal
2-5 ms
20-100 ms
(Depends on
Sample Size)
150 ms
A conservative planning estimate would be 10 ms for encryption delay and 10 ms for

decryption delay.
648
This delay might not seem signicant for a campus-to-branch call (hub-to-spoke), but
the delay might be more relevant in branch-to-branch (spoke-to-spoke) scenarios because
encryption and decryption might occur twice (depending on the logical topology of the
VPN). This is illustrated in Figure 16-8.
Figure 16-8 IPSec VPN Spoke-to-Spoke Encryption/Decryption Delays
Campus
(IPSec VPN Hub)
Decryption (2-5 ms)
Encryption (2-5 ms)
Branch A
(IPSec VPN Spoke A)
NOTE
Encryption (2-5 ms)
Decryption (2-5 ms)
Branch B
(IPSec VPN Spoke B)
Not only do encryption delays need to be factored into spoke-to-spoke IPSec VPN
scenarios, but queuing and serialization delays for both legs of the tunnel do as well
(as they also would in private WAN spoke-to-spoke scenarios).
ToS Byte Preservation

For the majority of QoS designs discussed thus far, classication is performed based on
DSCP markings in the ToS byte of the IP packet. However, when an IP packet is encrypted
649
through IPSec, the original ToS byte values also are encrypted and, thus, unusable by QoS
mechanisms that process the packet (post encryption).
To overcome this predicament, the IPSec protocol standards inherently have provisioned
the capability to preserve the ToS byte information of the original IP header by copying it
to the IP headers added by the tunneling and encryption process.
As shown in Figure 16-9, the original IP ToS byte values are copied initially to the IP header
added by the GRE encapsulation. Then these values are copied again to the IP header added
by IPSec encryption.
Figure 16-9 IP ToS Byte Preservation
ToS byte is copied from
the original IP header to
the IP GRE header.
ToS byte is copied again

from the IP GRE header
to the IPSec header.
GRE IP
GRE
Hdr

Hdr
Hdr
Hdr IV
GRE
IP
Hdr
UDP
RTP
Voice
IP
Hdr
UDP
RTP
Voice
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
This process compensates for the fact that the original IP header (including the ToS byte)
is actually unreadable (because of encryption) and allows the packet to be processed by
(post encryption) QoS mechanisms in the same manner as any other packet.
Additionally, this process underscores the importance of ensuring that the encrypted trafc
is marked properly (at Layer 3) before encryption.
QoS Pre-Classify
The QoS Pre-Classify feature often is confused with ToS byte preservation. QoS PreClassify is a Cisco IOS feature that allows for packets to be classied on header parameters
other than ToS byte values after encryption.
Because all original packet header elds are encrypted, including source or destination IP
addresses, Layer 4 protocol, and source or destination port addresses, post-encryption QoS
mechanisms cannot perform classication against criteria specied within any of these
elds.
650
A solution to this constraint is to create a clone of the original packets headers before
encryption. The crypto engine encrypts the original packet, and then the clone is associated
with the newly encrypted packet and sent to the output interface. At the output interface,
any QoS decisions based on header criteria, except for ToS byte valueswhich have been
preservedcan be performed by matching on any or all of the ve access-list tuple values
of the clone. In this manner, advanced classication can be administered even on encrypted
packets. The process is illustrated in Figure 16-10.
Figure 16-10 QoS Pre-Classify Feature Operation
Packet
clone
particle1
A clone of the original packet headers

is created and associated with the
encrypted particles.
particle2
Input Interface
particleN
%$#*&1
Crypto Engine
clone
QoS classification is performed

against the clone (as the original
packets headers are encrypted).
%$#*& 2
%$#*& N
class-map NET-MGMT
Output Interface
Encrypted Packet
!
access-list 123 permit tcp any host 10.45.15.1 eq telnet
QoS Pre-Classify allows for advanced classification (beyond ToS byte values)
such as source/dest IP addrs, Layer 4 protocols, and source/dest port numbers.
A key point to remember regarding QoS Pre-Classify is that it is applicable only at the
encrypting routers output interface. The elds preserved by QoS Pre-Classify are not
available to any routers downstream; the clone never leaves the router performing the
encryption, thus ensuring the integrity and security of the IPSec VPN tunnel.
QoS Pre-Classify is supported in all Cisco IOS switching paths and is recommended to be
enabled on some platforms even when only the ToS byte values are being used for
classication. Testing has shown that when hardware-based encryption cards are combined
with QoS, the Cisco IOS Software implementation of the QoS Pre-Classify feature slightly
enhances performance, even when matching only on ToS byte values. Furthermore,
enabling QoS Pre-Classify by default eliminates the possibility that its conguration will
be overlooked if the QoS policy later is changed to include matching on IP addresses, ports,
or protocols.
651
Design recommendations for the QoS Pre-Classify feature can be summarized as follows:
Enable QoS Pre-Classify on all branch IPSec VPN routers that support the feature.
Enable QoS Pre-Classify on headend IPSec VPN routers only when both the VPN
termination and QoS policies reside on the same device.
Pre-Encryption Queuing
The hardware crypto engine within a Cisco VPN routers chassis can be viewed as an
internal interface that processes packets for encryption or decryption.
Before Cisco IOS Release 12.2(13)T, packets to be encrypted were handed off to the crypto
engine in a rst-in-rst-out (FIFO) basis. No distinction was made between voice packets
and data packets. The FIFO queuing for crypto engines is illustrated in Figure 16-11.
Figure 16-11 FIFO Crypto Engine QoS
Input
Interface(s)
FIFO Input
Queue
Crypto
Engine
Output
Interface
Consider a Cisco 2651XM router deployed at a branch site congured with a full-duplex
Fast Ethernet interface, a Serial E1 interface (also full duplex), and an AIM-BP encryption
accelerator. The Fast Ethernet interface connects to the branchs LAN, and the serial
interface connects to the Internet. These factors could limit throughput (causing a
bottleneck) in this scenario:
The clock rate of the slowest interface (in bits per secondin this case, 2 Mbps
transmitted or 2 Mbps received over the E1 interface)
The packet-forwarding rate of the routers main CPU (in packets per second)
The crypto engine encryption/decryption rate (in packets per second)
The performance characteristics of these items further are inuenced by the trafc mix,
including the rates and sizes of the IP packets being switched through the router and the
congured Cisco IOS switching path (process-switched, fast-switched, or CEF-switched).
NOTE
In most hardware platforms, the packets-per-second capabilities of the router are more
important for planning purposes than bits per second switched through the router. For
example, if the average packet size of packets switched through the router increases from
128 bytes to 256 bytes, the packet-per-second capabilities of the main CPU are not
necessarily cut in half.
652
The control plane requirements also factor into the CPUs utilization. These requirements
are determined by the routing protocol(s) in use, the number of routes in the routing table,
the overall network stability, and any redistribution requirements. Management requirements such as NTP and SNMP also add to the CPU tax. Additionally, Cisco IOS HA, QoS,
multicast, and security features all consume CPU resources and must be taken into account.
Another factor in the equation is the ratio of packets switched through (and originated by)
the router in relation to the number of packets selected by the crypto maps access list for
encryption or decryption. If an IP GRE tunnel is being encrypted, this tends to be a large
percentage of encrypted packets to total packets; if an IP GRE tunnel is not being
encrypted, the ratio could be quite small.
Hardware crypto engines can become congested when their packet-processing capabilities
are less than those of the routers main CPU and interface clock speeds. In such a case, the
crypto engine becomes a bottleneck, or a congestion point. The crypto engine might be
oversubscribed on either a momentary or (worse case) sustained basis. Such internal
precrypto congestion could affect the quality of real-time applications, such as VoIP.
Cisco internal testing and evaluation has shown it to be extremely difcult for conditions to
arise that cause hardware crypto engine congestion. In nearly all cases, the Cisco VPN
router platforms main CPU is exhausted before reaching the limit of the crypto engines
packet-processing capabilities.
Nevertheless, Cisco provides a solution to this potential case of congestion in the rare event
that a hardware crypto engine is overwhelmed so that VoIP quality will be preserved. This
feature, low-latency queuing (LLQ) for IPSec encryption engines, was introduced in Cisco
IOS Release 12.2(13)T.
The LLQ for Crypto Engine feature provides a dual-input queuing strategy for packets
being sent to the crypto engine:
A priority or low-latency queue

A best-effort queue
This feature is targeted at alleviating any effects of momentary or sustained oversubscription of the hardware crypto engines that could result in priority trafc (such as voice and
video) experiencing quality issues. This feature is illustrated in Figure 16-12.
NOTE
Because software-based crypto adds unacceptable latency and jitter, there are no plans to
incorporate this feature for software crypto. Hardware accelerators for IPSec encryption are
highly recommended.
653
Figure 16-12 LLQ (Dual-FIFO) Crypto Engine QoS
Input
Interface(s)
LLQ/CBWFQ
Classification
LLQ
Best Effort
Crypto
Engine
Output
Interface
The classication component to segregate trafc between the priority (LLQ) queue and the
best-effort queue is based on the MQC service policy on the output interface(s).
No additional conguration is required to enable LLQ for crypto engines; it is enabled
internally by the presence of a service policy with an LLQ priority command that is
applied to an output interface of an IPSec VPN router.
Trafc specied in the service policy to be assigned to the interfaces priority queue (LLQ)
automatically is sent to the crypto engines LLQ. Trafc included in any CBWFQ bandwidth classes (including the default class) automatically is assigned to the crypto engines
best-effort queue.
It is possible to congure different service policies, each with different trafc assigned to
an LLQ, on different interfaces. For example, perhaps voice is assigned to the LLQ of
Serial1/0 and video is assigned to the LLQ of an ATM PVC. Assuming that both voice and
video are to be encrypted, the question arises, which type of trafc (voice or video) will be
assigned to the crypto engines LLQ?
Because the crypto engine acts like a single interface inside the VPN router, encrypting and
decrypting all outbound and inbound trafc streams for each interface on which crypto is
applied, in the case of multiple service policies (on different interfaces) the crypto engine
maps all interface priority queues (LLQ) to its LLQ and all other queues to its best-effort
queue. Therefore, both voice and video would be assigned to the crypto engines LLQ.
In short, the LLQ for Crypto Engine feature ensures that if packets are dropped by momentary or sustained congestion of the crypto engine, the dropped packets will be of appropriately lower priority (not VoIP packets).
Although the feature is enabled by the presence of a service policy with an LLQ priority
statement, as with interface queuing itself, crypto-engine queuing does not actually engage
prioritization through the dual-FIFO queuing strategy until the crypto engine itself
experiences congestion.
654
The LLQ for Crypto Engine feature in Cisco IOS Software is not a prerequisite for
deploying QoS for IPSec VPN implementations in a high-quality manner. As indicated
previously, internal Cisco evaluations have found it extremely difcult to produce network
trafc conditions that resulted in VoIP quality suffering because of congestion of the
hardware crypto engine.
In general, the LLQ for Crypto Engine feature offers the most benet under one of the
following conditions:
When implementing Cisco IOS VPN router platforms that have a relatively high
amount of main CPU resources relative to crypto engine resources (these vary
depending on the factors outlined earlier in this discussion).
When the network experiences a periodic or sustained burst of large packets (for
example, for video applications).
To summarize, high-quality IPSec VPN deployments are possible today without the LLQ
for Crypto Engine feature in Cisco IOS software. The addition of this feature in Cisco IOS
software further ensures that high-priority applications, such as voice and video, can operate
in a high-quality manner even under harsh network conditions.
Anti-Replay Implications
IPSec offers inherent message-integrity mechanisms to provide a means to identify whether
an individual packet is being replayed by an interceptor or hacker. This concept is called
connectionless integrity. IPSec also provides for a partial sequence integrity, preventing the
arrival of duplicate packets. These concepts are outlined in RFC 2401, Security Architecture
for the Internet Protocol.
When ESP authentication (esp-sha-hmac) is congured in an IPSec transform set, for each
security association, the receiving IPSec peer veries that packets are received only once.
Because two IPSec peers can send millions of packets, a 64-packet sliding window
is implemented to bound the amount of memory required to tally the receipt of a peers
packets. Packets can arrive out of order, but they must be received within the scope of the
window to be accepted. If they arrive too late (outside the window), they are dropped.
The operation of the Anti-Replay window protocol is as follows:
1 The sender assigns a unique sequence number (per security association) to encrypted
packets.
2 The receiver maintains a 64-packet sliding window, the right edge of which includes
the highest sequence number received. In addition, a Boolean variable is maintained

to indicate whether each packet in the current window was received.
655
3 The receiver evaluates the received packets sequence number:
If a received packets sequence number falls within the window and was not
received previously, the packet is accepted and marked as received.
If the received packets sequence number falls within the window and
previously was received, the packet is dropped and the replay error counter
is incremented.
If the received packets sequence number is greater than the highest
sequence in the window, the packet is accepted and marked as received, and
the sliding window is moved to the right.
If the received packets sequence number is less than the lowest sequence in
the window, the packet is dropped and the replay error counter is
incremented.
In a converged IPSec VPN implementation with QoS enabled, lower-priority packets are
delayed so that higher-priority packets receive preferential treatment. This has the unfortunate side effect of reordering the packets to be out of sequence from an IPSec Anti-Replay
sequence number perspective. Therefore, there is a concern that through the normal QoS
prioritization process, the receiver might drop packets as Anti-Replay errors, when, in fact,
they are legitimately sent or received packets.
Figure 16-13 provides a visualization of the process. In this example, voice packets 4
through 67 have been received, and data packet 3 was delayed and transmitted following
voice packet 68. When the Anti-Replay logic is called to process packet 3, it is dropped
because it is outside the left edge of the sliding window. Packets can be received out of
order, but they must fall within the window to be accepted.
Figure 16-13 Anti-Replay Operation
Outside
Window
64-Packet
Anti-Replay
Window
64 Packet
SlidingSliding
Window
69
68
3
Anti-Replay
Drop
64
65
66
67
656
Anti-Replay drops can be eliminated in a pure IPSec tunnel design (no encrypted IP GRE
tunnel) by creating separate security associations for voice and data; voice and data packets
must match a separate line in the access list referenced by the crypto map. This is implemented easily if the IP phones are addressed by network addresses (such as private RFC
1918 addresses) separate from the workstations.
However, if IPSec tunnel mode (with an encrypted IP GRE tunnel) is used for a converged
network of voice and data, Anti-Replay drops impact data packets instead of voice packets
because the QoS policies prioritize voice over data.
Consider the effect of packet loss on a TCP-based application: TCP is connection oriented
and incorporates a ow-control mechanism within the protocol. The TCP application
cannot see why a packet was dropped. A packet dropped by a service policy on a congested
output interface is no different to the application than a packet lost by an Anti-Replay drop.
From a network perspective, however, it would be more efcient to drop the packet before
sending it over the WAN link (where bandwidth is the most expensive, only to have it
dropped by the Anti-Replay mechanism on the receiving IPSec VPN router), but the
location or nature of the packet loss is immaterial to the TCP driver.
Anti-Replay drops of data trafc ows are usually in the order of 1 percent to 1.5 percent
on IPSec VPN links that experience sustained congestion and have queuing engaged
(without any additional policy tuning).
Output drops on the output WAN interface, however, tend to be few, if any, and certainly
are far fewer than those dropped by Anti-Replay. This is because Anti-Replay triggers
packet drops more aggressively than the output service policy, which is a function of the
size of the output queues and the number of dened classes.
By default, each CBWFQ class receives a queue with a length of 64 packets. This can be
veried with the show policy interface verication command. Meanwhile, the receiving
IPSec peer has a single 64-packet Anti-Replay window (per IPSec Security Association)
with which to process all packets from all LLQ and CBWFQ bandwidth classes.
So, it stands to reason that the Anti-Replay mechanism on the receiving VPN router will be
more aggressive at dropping packets delayed by QoS mechanisms preferential to VoIP than
the service policy at the sending router. This is because of the size mismatch of the queue
depth on the senders output interface (multiple queues of 64 packets each) compared to the
width of the receivers Anti-Replay window (a single sliding window of 64 packets per SA).
As more bandwidth classes are dened in the policy map, this mismatch increases. As
mentioned, this is an inefcient use of expensive WAN/VPN bandwidth because packets
are transmitted only to be dropped before decryption.
The default value of 64 packets per CBWFQ queue is designed to absorb bursts of data
trafc and delay rather than drop those packets. This is optimal behavior in a non-IPSec
enabled network.
657
When IPSec authentication is congured (esp-sha-hmac) in the network, the scenario can
be improved by reducing the queue limit (max threshold) of the bandwidth classes of the
senders output service policy so that it becomes more aggressive at dropping packets than
buffering or delaying them. Extensive lab testing has shown that such queue-limit tuning
can reduce the number of Anti-Replay drops from 1 percent to less than a tenth of percent
(< 0.1 percent). This is because decreasing the service policy queue limits causes the senders
output service policy to become more aggressive in dropping instead of signicantly delaying packets (which occurs with large queue depths). This, in turn, decreases the number of
Anti-Replay drops.
As a rule of thumb, the queue limits should be reduced in descending order of application
priority. For example, the queue limit for Scavenger should be set lower than the queue limit
for a Transactional Data class, as is shown later in Examples 16-3 through 16-6.
NOTE
In many networks, the default queue limits and IPSec Anti-Replay performance are considered
acceptable. A modication of queue-limit values entails side effects on the QoS service
policy and related CPU performance. When queue limits are tuned, a careful eye should be
kept on CPU levels.
Control Plane Provisioning

As discussed in Chapter 13, Cisco IOS Software has an internal mechanism for protecting
control trafc, such as routing updates, called PAK_priority.
PAK_priority marks routing protocols to DSCP CS6, but it currently does not protect IPSec
control protocols, such as ISAKMP (UDP port 500). Therefore, it is recommended to
provision an explicit CBWFQ bandwidth class for control plane trafc, including ISAKMP,
as shown in Example 16-2.
Example 16-2 Protecting IPSec Control Plane Trafc
!
class-map match-any INTERNETWORK-CONTROL
match ip dscp cs6
match access-group name IKE
!
!
policy-map V3PN
...
class INTERNETWORK-CONTROL
bandwidth percent 5
...
!
ip access-list extended IKE
permit udp any eq isakmp any eq isakmp
!
! References ISAKMP ACL
! Control Plane provisioning
! ISAKMP ACL
658
Site-to-Site V3PN QoS Designs

As with WAN and MPLS VPN QoS models, site-to-site V3PN QoS models can range from
a basic number of classes (in this case, the minimum number of recommended classes is 6)
to a complex QoS Baseline model (11 classes). Each enterprise must determine present
needs and comfort level of QoS complexity, along with future needs, to more easily migrate
to progressively more complex QoS models, as required.
Six-Class Site-to-Site V3PN Model

The six-class V3PN model technically should be referred to as V2PN because it includes
provisioning for only voice (not for video) over an IPSec VPN. Voice is protected explicitly
with LLQ; call signaling and control plane trafc also explicitly are protected through
CBWFQ classes. This model includes a preferential data class (Transactional Data) and
a deferential class (Scavenger, which is squelched to the minimum congurable amount:
1 percent). If the queue limits are tuned to minimize Anti-Replay drops, the queue limit for
Transactional Data should be set higher than the queue limit for class default.
An example six-class V3PN model, which is suitable for link speeds up to and including
T1/E1 speeds, is illustrated in Figure 16-14 and detailed in Example 16-3.
Figure 16-14 Six-Class Site-to-Site V3PN Model
Best-Effort
25%
Voice
33%
Scavenger 1%
Call-Signaling 5%
Transactional Data 31%
Internetwork Control 5%
659
Example 16-3 Six-Class Site-to-Site V3PN Model Conguration Example

!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
match ip dscp cs6
match ip dscp cs1
!
!
policy-map SIX-CLASS-V3PN-EDGE
class VOICE
priority percent 33
bandwidth percent 5
bandwidth percent 5
queue-limit 20
class SCAVENGER
bandwidth percent 1
queue-limit 1
class class-default
queue-limit 16
!
!
!
NOTE
! VoIP
! New Call-Signaling
! Old Call-Signaling
! IP Routing
! Transactional-Data
! Scavenger
! VoIP gets 33% LLQ

! Call-Signaling provisioning
! Transactional-Data provisioning
! Optional: Anti-Replay tuning
! Best Effort needs BW guarantee
! Optional: Anti-Replay Tuning
! ISAKMP ACL
Currently, only distributed platforms support the queue-limit command in conjunction

with WRED commands. However, this command combination will be available on
nondistributed platforms with the release of Consistent QoS Behavior, as discussed in
Chapter 13.
660
show policy
Eight-Class Site-to-Site V3PN Model

With the addition of a class for Interactive-Video, this model more accurately lives up to its
V3PN name. Interactive-Video (as with VoIP) must be provisioned adequately to include
IPSec encryption overhead, but (unlike VoIP) there are no clean formulas for calculating
the required incremental bandwidth. This is because video packet sizes and packet rates
vary signicantly and are largely a function of the degree of motion within the video images
being transmitted.
On an unencrypted topology, the guideline for provisioning for Interactive-Video is to
overprovision the LLQ by 20 percent. Although this conservative guideline holds true
in most videoconferencing scenarios, which usually consist of relatively minor motion
(meetings, conferences, talking heads, and so on), in some cases this rule proves inadequate
over an IPSec VPN. In such cases, the LLQ might need to be provisioned to the streams
rate plus 25 percent or more. The need to increase the bandwidth for Interactive-Videos
LLQ will be apparent if drops (in excess of 1 percent) appear under this class when using
the verication command show policy interface.
The other class added to this model is a separate class for Bulk Data applications (which
will be constrained if congestion occurs, to prevent long sessions of TCP-based ows from
dominating the link). Notice that the queue limit of the Bulk Data class has been reduced
to below the queue limit of the Best-Effort class. This is because of the relative ranking of
application priority: During periods of congestion, the Bulk Data class (large TCP-based
le operations that operate mainly in the background) is prevented from dominating
bandwidth away from the Best-Effort class (as a whole).
This policy is suitable for link speeds of 3 Mbps or higher. However, in an IPSec environment, it is good to remember that load-sharing GRE tunnels over multiple physical interfaces
exacerbate Anti-Replay drops. Whenever possible, a single physical interface should be
used to achieve these higher speeds. Another consideration to bear in mind is that higher-end
platforms, such as the 2691 and 3700- or 7200-series routers, are required to perform crypto
at higher speeds. The Eight-Class Site-to-Site V3PN model is illustrated in Figure 16-15
and detailed in Example 16-4.
661
Figure 16-15 Eight-Class Site-to-Site V3PN Model

Voice
18%
Best-Effort
25%
Interactive-Video
15%
Scavenger 1%
Bulk Data 4%
Call-Signaling 5%
Example 16-4 Eight-Class Site-to-Site V3PN Model Conguration Example

!
class-map match-all
match ip dscp ef
class-map match-all
match ip dscp af41
class-map match-any
match ip dscp cs3
match ip dscp af31
class-map match-any
match ip dscp cs6
match access-group
class-map match-all
match ip dscp af21
class-map match-all
match ip dscp af11
class-map match-all
match ip dscp cs1
VOICE
! VoIP
INTERACTIVE-VIDEO
af42
CALL-SIGNALING
! Interactive-Video
INTERNETWORK-CONTROL
name IKE
TRANSACTIONAL-DATA
af22
BULK-DATA
af12
SCAVENGER
! IP Routing
! Bulk Data
! Scavenger
!
policy-map EIGHT-CLASS-V3PN-EDGE
class VOICE
priority percent 18
priority percent 15
! VoIP gets 18% LLQ

! IP/VC gets 15% LLQ
continues
662
Example 16-4 Eight-Class Site-to-Site V3PN Model Conguration Example

bandwidth percent 5
bandwidth percent 5
queue-limit 18
class BULK-DATA
bandwidth percent 4
queue-limit 3
class SCAVENGER
bandwidth percent 1
queue-limit 1
class class-default
queue-limit 16
! Bulk-Data provisioning
!
!
! ISAKMP ACL
show policy
QoS Baseline (11-Class) Site-to-Site V3PN Model

Building on the previous model, three new classes are added: Network-Management,
Mission-Critical Data, and Streaming-Video.
This model also is suitable only for high-speed (3 Mbps and above) links, with the same
caveats regarding multiple physical links aggravating Anti-Replay drops and the requirement of using newer platforms to perform crypto at these speeds. The queue limits have
been tuned to reect relative application priority.
The QoS Baseline V3PN model, suitable for 3-Mbps link speeds and higher, is illustrated
in Figure 16-16 and detailed in Example 16-5.
663
Figure 16-16 QoS Baseline (Eleven-Class) Site-to-Site V3PN Model

Best-Effort
25%
Voice 18%
Interactive-Video
15%
Scavenger 1%
Bulk Data 4%
Streaming-Video
10%
Call-Signaling 5%
Example 16-5 QoS Baseline (Eleven-Class) Site-to-Site V3PN Model Conguration Example
!
class-map match-all
match ip dscp ef
class-map match-all
match ip dscp af41
class-map match-any
match ip dscp cs3
match ip dscp af31
class-map match-any
match ip dscp cs6
match access-group
class-map match-all
match ip dscp cs2
class-map match-all
match ip dscp 25
class-map match-all
match ip dscp af21
class-map match-all
match ip dscp cs4
class-map match-all
match ip dscp af11
class-map match-all
match ip dscp cs1
VOICE
! VoIP
INTERACTIVE-VIDEO
af42
CALL-SIGNALING
! Interactive-Video
name IKE
NET-MGMT
! IP Routing
! Network Management
MISSION-CRITICAL-DATA
! Interim MC Data
TRANSACTIONAL-DATA
af22
STREAMING-VIDEO
! Transactional Data
! Streaming Video
BULK-DATA
af12
SCAVENGER
! Bulk Data
! Scavenger
continues
664
Example 16-5 QoS Baseline (Eleven-Class) Site-to-Site V3PN Model Conguration Example (Continued)
!
policy-map QOSBASELINE-V3PN-EDGE
class VOICE
priority percent 18
! VoIP gets 18% LLQ
priority percent 15
! IP/VC gets 15% LLQ
bandwidth percent 5
bandwidth percent 5
class NET-MGMT
bandwidth percent 2
! Network Management provisioning
! Mission-Critical Data provisioning
queue-limit 6
bandwidth percent 5
queue-limit 3
! Streaming-Video provisioning
queue-limit 6
class BULK-DATA
bandwidth percent 4
! Bulk-Data provisioning
queue-limit 3
class SCAVENGER
bandwidth percent 1
! Scavenger throttling
queue-limit 1
class class-default
queue-limit 16
!
!
! ISAKMP ACL
!
show policy
At remote sites, these policies can be applied to the physical interfaces connecting them to
the service provider (provided that the SLAs are for line ratesotherwise, shapers must be
used). For example, if the service provider guarantees a full T1 rate to a remote site and the
access medium is Frame Relay, the service policy can be applied directly to the main Frame
Relay interface. If the service provider guarantees only 768 kbps across the same link,
Frame Relay trafc shaping (either legacy or class-based FRTS) combined with FRF.12
must be used at the remote site.
Headend VPN Edge QoS Options for Site-to-Site V3PNs
665
For the central site(s) WAN aggregators, however, unique considerations exist. These are
discussed in the next section.
Headend VPN Edge QoS Options for Site-to-Site V3PNs

IPSec V3PNs can be congured in various ways at the central sites. Some enterprises
simply overlay V3PNs on top of their existing private WANs; others subscribe to service
providers that offer classes of service within their clouds. Many enterprises deploy VPN
headends behind WAN aggregation routers to distribute CPU loads, while some perform
encryption and QoS on the same box. Each of these options presents considerations on how
V3PN policies optimally are applied on WAN aggregation routers.
For enterprises that have overlaid IPSec VPNs on top of their private WAN topologies,
the V3PN policies should be applied to the leased lines or Frame Relay/ATM PVCs, as
described in Chapter 13.
For enterprises that are subscribing to service providers that offer PE-to-CE QoS classes
(including enterprises that are deploying IPSec VPNs over MPLS VPNs), V3PN policies
need to be applied on the CE-to-PE links (complete with any re-marking that the service
provider requires to map into these service provider classes), as described in Chapter 15,
MPLS VPN QoS Design.
For enterprises that are subscribing to service providers that do not offer explicit QoS
(beyond an SLA) within their cloud and are using VPN headends behind WAN aggregation
routers, the V3PN service policies would be applied to the WAN aggregators (CE-to-PE)
physical links. This prioritizes packets (by applications) and relies on the service providers
SLA to ensure that the delivery largely reects the priority of the packets as they are handed
off to the service provider. Such a conguration would require only a single QoS policy for
a WAN aggregator (albeit, on a high-speed interface), but at the same time, it would involve
an increased dependence on the service providers SLA to deliver the desired QoS service
levels.
When the VPN headend routers have adequate CPU cycles to perform QoS, another option
exists: hierarchical MQC policies that shape and queue (within the shaped rate) and are
applied on a per-tunnel basis. A sample of per-tunnel hierarchal QoS policies is shown in
Example 16-6. As with previous shaping design recommendations, the shaper is congured
to shape to 95 percent of the remote sites line rate.
NOTE
It is critical to keep an eye on CPU levels when IPSec VPN encryption and per-tunnel
QoS policies are applied on the same router. CPU levels, in general, should not exceed
75 percent during normal operating conditions. Conguring hierarchical shaping and
queuing policies on a per-tunnel (per-SA) basis to a large number of sites could be very
CPU intensive, especially when such sites experience periods of sustained congestion.
666
Example 16-6 Per-Tunnel Hierarchical Shaping and Queuing MQC Policy for VPN Headends/WAN Aggregators
!
policy-map SHAPING-T1-TUNNEL
class class-default
! Shaped to 95% of T1 line-rate
service-policy V3PN-EDGE
! Nested queuing policy
!
!
interface Tunnel120
description VPN Pipe to V3PN Site#120 (T1 Link)
bandwidth 1536
ip address 10.10.120.1 255.255.255.252
ip mtu 1420
service-policy output SHAPING-T1-TUNNEL
! Policy applied to tunnel int
qos pre-classify
! Performance recommandation
tunnel source 192.168.1.1
crypto map VPN
!
Teleworker V3PN QoS Considerations

Organizations constantly are striving to reduce costs and improve productivity and
employee retention. Teleworker solutions address these organizational objectives by
giving employees the ability to work from home with compatible quality, functionality,
performance, convenience, and security, as compared to working in an ofce location.
Teleworker solutions that provide such functionality have been branded enterprise class
or business ready by Cisco Marketing and are illustrated in Figure 16-17.
Figure 16-17 Business-Ready Teleworker Design
IP Telephony
Infrastructure
VPN Headend
Routers
Central Site
IPSec
Tunnel
Home VPN
Router
Service
Provider
Internet
Broadband
Provider
Home Office
667
Telecommuter solutions include these main benets:
Increased productivityOn average, employees spend 60 percent of their time or

less at their desks, yet this is where the bulk of investment is made in providing access
to corporate applications. Providing similar services at an employees residence, for a
relatively minor investment, signicantly can increase productivity gains.
Business resilienceEmployees can be displaced from their normal workplace by

natural events (such as winter storms, hurricanes, or earthquakes), health alerts (such
as SARS), man-made events (such as travel restrictions or trafc conditions), or simply family-related events, such as sick children or home repairs. These disruptions signicantly can impact an organizations processes. Providing employees with central
siteequivalent access to applications and services in geographically dispersed locations (such as home ofces) creates a built-in back-up plan to keep business processes
functioning in unforeseen circumstances.
Cost savingsA traditional remote worker setup involves toll charges for dial-up and
additional phone lines. Integrating services into a single, broadband-based connection
can eliminate these charges while delivering superior overall connectivity
performance.
SecurityDemands for access to enterprise applications outside the campus

are stretching the limits of security policies. Teleworking over IPSec VPNs offers
inherent security provided by encryption of all trafc, including data, voice, and
video. Also critical is integrating rewall and intrusion-detection capabilities, along
with nding ways to easily accommodate both corporate and personal users who share
a single broadband connection (the spouse-and-children concern, which will be
discussed shortly).
Employee recruitment and retentionIn the past, enterprises recruited employees

in the locations where corporate ofces were located. Today enterprises need the
exibility of hiring skilled employees wherever the skills exist and need to integrate
remote workers into geographically dispersed teams with access to equivalent
corporate applications.
Although QoS designs for IPSec V3PN teleworker scenarios share many of the same
concerns of site-to-site V3PN scenarios, additional specic considerations relating to
teleworker deployment models and broadband access technologies need to be taken into
account.
Teleworker Deployment Models

Business-ready teleworker deployment models must provide the following services:
Basic servicesThese include NAT, DHCP, IP routing, and multiple Ethernet

connections for home ofce devices and broadband connection (attachment to the
WAN circuit cable, DSL, ISDN, or wireless network).
668
QoS servicesThese include classication, prioritization, and, in some cases,

shaping of trafc.
VPN/security servicesThese include encryption of trafc to the main site and

rewall functionality for the home ofce.
Given these requirements, there are three main deployment models for business-ready
teleworker V3PN solutions, as illustrated in Figure 16-18. These include the Integrated
Unit Model, the Dual-Unit Model, and the Integrated Unit + Access Model.
Figure 16-18 Business-Ready Teleworker Deployment Models
Integrated Unit Model
IP
Router
QoS, VPN
Dual Unit Model
Service
Provider
Backbone
IP
VPN
Device
Integrated Unit +
Access Model
IP
Router
QoS, VPN
Enterprise
Router
QoS
Broadband
Access
Device

In the Integrated Unit Model, a single device (such as a Cisco 837 or 1700-series router)
provides basic services, QoS services, and VPN/security services. Furthermore, the device
connects directly to the broadband media.
NOTE
The Cisco routers used in these scenarios require an IP/FW/PLUS 3DES Cisco IOS
Software feature set. This feature set would include support for LLQ/CBWFQ with classbased hierarchical trafc shaping, and also support for Easy VPN (EZVPN), PKI, IDS,
AES, URL ltering, and EIGRP.
Advantages include the following:
Single-device deployment and management
Potential cost savings
Adaptability for service provider fully managed services (transport, QoS, IP

Telephony application)
669
Disadvantages include the following:
Availability of a single device at an appropriate cost with the features and

performance required
No single unit for some broadband access circuit types
The Integrated Model is a preferred model for service providers offering a fully managed
V3PN teleworker service.
From a QoS design perspective, this model is highly similar to a site-to-site V3PN, except
that it interfaces with a broadband media instead of a traditional WAN access media.
Dual-Unit Model
In this model, one device (such as a Cisco 831, 837, or 1700-series router) provides basic
services and QoS, and a second device (a Cisco router or a PIX 501 or even a VPN 3002)
performs security/VPN services.
Granularity of managed servicesService providers can manage broadband access

while enterprises manage VPN/security and private network addressing because these
are two different units.
Media independenceBecause the VPN/security device is separate from the router

connecting to the broadband circuit, the same VPN device can be used for cable, DSL,
wireless, and ISDN by changing the router model or module in the router. This is
especially valuable if one enterprise must support teleworkers with different broadband circuit types (such as DSL, cable, and ISDN).
Two units packaged for deployment

Ongoing management of two devices
The cost for two devices
From a QoS design perspective, this model is no different from the previous (Integrated
Unit) model.
Integrated Unit + Access Model

In this third model, a single router (such as a Cisco 831 or 1700-series router) provides basic
services, QoS services, and VPN/security services. However, the router does not connect
directly to the broadband access media; it connects (through Ethernet) to a broadband
access device (such as a DSL or cable modem).
670
Cost savings realized by using existing broadband-access devices
Solution support even when no router interface for a specic broadband circuit type
is available
Simplied provisioning because one size ts all, regardless of broadband access

media
Increased troubleshooting complexity because most broadband-access devices

(modem) are not intelligent and, therefore, cannot be queried, managed, or controlled.
Additional QoS complexity because, in this model, the router does not control the
broadband circuit and, thus, must perform hierarchical shaping, as shown in
Figure 16-19.
Figure 16-19 Hierarchical QoS Requirements for Integrated Unit + Access Teleworker Deployment Model
Logical Bottleneck: Traffic is shaped to uplink speed on Ethernet egress of
the VPN/QoS router to force queueing to engage at this
interface whenever the uplink speed is exceeded.
10/100 Mbps
10 Mbps
384 kbps
Ethernet/Fast Ethernet
Ethernet
VPN/QoS
Router
Broadband
Modem
Broadband
Service
Provider
Internet
Service
Provider
VPN
Headend
Physical Bottleneck:
Without induced shaping, congestion would
occur at broadband modem, but this device
lacks QoS capabilities.
Because of the media-specic encapsulation overhead requirements (discussed in the

following sections), it is recommended to shape to 95 percent of broadband link for cable
and 70 percent of the uplink rate for DSL.
A hierarchical shaping policy that forces queuing to engage for a 384-kbps cable broadband
connection is shown in Example 16-7.
671
Example 16-7 Hierarchical Shaping and Queuing MQC Policy for a 384-kbps Cable Connection
!
policy-map SHAPE-384-CABLE
class class-default
shape average 364800 3640
! Shapes to 95% of 384 kbps cable link
service-policy V3PN-TELEWORKER
! Nested V3PN Teleworker queuing policy
!
...
!
interface Ethernet0
service-policy output SHAPE-384-CABLE
! Shaper applied to LAN interface
!
The Integrated Unit + Access Model is a preferred model for enterprise V3PN teleworker
deployments because it completely abstracts any service provider or access media
specic requirements (a one-size-ts-all solution).
Broadband-Access Technologies
In North America, there are four main broadband-access technology options for telecommuter scenarios: DSL, cable, ISDN, and wireless. DSL and cable are by far the dominant
broadband technologies.
Because of per-minute costs, ISDN is used considerably less; however, ISDN at rate is
becoming available and will make ISDN a good option for areas where DSL or cable is
not available. Last-mile wireless is a new option that is being piloted in certain areas to
determine viability.
The minimum recommended broadband data rate for most deployments is 160 kbps
(uplink)/860 kbps (downlink). Data rates below this speed require more troubleshooting by
support staff and are less likely to provide acceptable voice quality. The recommended data
rate for V3PN teleworker deployments is 256 kbps (uplink)/1.4 Mbps (downlink) or higher
rates. Although V3PN can be deployed at rates less than 160 kbps/860 kbps, generally the
voice quality at that service level is in the cell phone quality range, and support costs are
higher.
Because QoS design for ISDN was discussed in Chapter 13, and because wireless as a lastmile broadband technology has yet to gain wide deployment, this section focuses only on
DSL and cable broadband technologies.
DSL and cable topologies are illustrated in Figure 16-20. Cable is a shared medium
between the teleworkers cable modem and the broadband providers cable headend router.
DSL is a dedicated circuit between the teleworkers DSL modem (bridge) and the DSL
Access Multiplexer (DSLAM). Both cable and DSL offerings utilize shared uplinks
between these aggregation devices and the service providers core network. QoS typically
672
is not provisioned within the broadband service providers cloud in either medium. This
lack of QoS within the broadband cloud underscores the requirement of adequate QoS
provisioning at the endpoints of the VPN tunnels.
Figure 16-20 DSL and Cable Topologies
IPSec V3PN
Headend
Routers
ATM
ATM
DSLAM/IP
DSL Switch
DSL
Bridge
Cable
Modem
IPSec V3PN
Teleworker
Routers
Broadband
Service Provider
Broadband Router
DSL (PPPoE) and cable (DOCSIS 1.0) in

service-provider networks are BestEffort with no QoS or serialization mitigation.
Digital Subscriber Line

DSL service features a dedicated access circuit and offers a service similar to Frame Relay
or Asynchronous Transfer Mode (ATM), in which a single permanent virtual circuit (PVC)
is provisioned from the home ofce to the service provider aggregation point. DSL has a
variety of speeds and encoding schemes.
Most service providers today offer residential Asynchronous Digital Subscriber Line
(ADSL). ADSL provides for asymmetric speeds (with the downstream rate greater than the
upstream rate). Asymmetrical links are a better choice and benet for the telecommuter
because the greater the downlink bandwidth is, the less the need is for QoS. (Slower-speed
uplinks slow TCP sender transmission rates to more manageable levels.) Because QoS is
not generally available by broadband service providers, such uplink/downlink speed
mismatches are desirable.
673
Residential access speeds are generally 128 to 384 kbps upstream and 608 kbps to 1.5 Mbps
downstream. For ADSL circuits with upstream speeds less than 256 kbps, G.729 VoIP
codecs are recommended.
ADSL also utilizes a single best-effort PVC using RFC 2516based PPP over Ethernet
(PPPoE) encapsulation. In DSL networks, delay and jitter are very low but are not guaranteed. Because PPPoE is used, no LFI is available in service provider DSL networks. In the
future, QoS at Layer 2 might be available across service provider networks using familiar
ATM variable bit-rate (VBR) denitions.
Single-pair high bit-rate DSL (G.SHDSL) is the new high-speed standard for business
DSL. Most residences will continue to be served by ADSL, while small business and
branch ofces likely will use G.SHDSL. G.SHDSL offers varying rates controlled by the
service provider; the upstream and downstream rates are the same speed (symmetric).
G.SHDSL is seen as an eventual replacement for T1s in the United States and will become
increasingly available from more service providers.
The telecommuter deployment options for DSL circuits include all three teleworker
deployment models: Integrated Unit, Dual-Unit, and Integrated Unit + Access, as shown
in Figure 16-21.
Figure 16-21 Teleworker Deployment Models for DSL

IP
Router
QoS, VPN
Dual Unit Model
DSL
Backbone
IP
To Enterprise
VPN
Device
Integrated Unit +
Access Model
IP
831/171X
DSL
Modem
Cable
Cable offers a shared service with symmetric speeds varying from 100 kbps to 4 Mbps. In
the past, delay and jitter varied too greatly over cable, making it unsuitable for VoIP.
The common installed base of cable services today is made up of Data-over-Cable Service
Interface Specications (DOCSIS) 1.0. No LFI is available with DOCSIS 1.0, although
DOCSIS 1.1 denes fragmentation and interleaving mechanisms.
674
DOCSIS 1.1 also provides the capabilities to shape trafc at Layer 2 before transmission
over the cable network. Although the circuit and frequencies physically are shared, access
to the medium can be controlled by the headend so that a device can be guaranteed specied
bandwidth.
At the time of this writing, the only recommended deployment model for cable is the
Integrated Unit + Access Model, as shown in Figure 16-22.
Figure 16-22 Teleworker Deployment Model for Cable
Integrated Unit +
Access Model
DSL
Backbone
To Enterprise
IP
831/171X
DSL
Modem
A few key differences with respect to bandwidth provisioning need to be taken into account
for broadband teleworker scenarios, as compared to site-to-site VPNs.
The rst is that usually only a single call needs to be provisioned for (unless two teleworkers share a single residential broadband connection, which is rarely the case). If bandwidth
is low, G.729 codecs are recommended. The second key bandwidth-provisioning consideration is the inclusion of the overhead required by the broadband access technologies,
whether DSL or cable.
NOTE
Sometimes multicast support is not required in a teleworker environment. For example,

routing protocols (which depend on multicast support) might not be required by teleworkers (because default gateways are usually sufcient in this context). In such cases, IPSec
tunnel mode (no encrypted IP GRE tunnel) can be used to achieve additional bandwidth
savings of 24 bytes per packet.
For the remainder of this chapter, IPSec tunnel mode (no encrypted IP GRE tunnel) is the
assumed mode of operation.
NAT Transparency Feature Overhead

Beginning in Cisco IOS Release 12.2(13)T, NAT transparency was introduced and is
enabled by default, provided that both peers support the feature. It is negotiated in the IKE
675
exchange between the peers. This feature addresses the environment in which IPSec
packets must pass through a NAT/pNAT device, and it adds 16 bytes to each voice and data
packet. The overhead that this feature adds is shown in Figure 16-23.
Figure 16-23 NAT Transparency Feature (Layer 3) Overhead
IPSec ESP Tunnel Mode UDP Encapsulation 128 Bytes
IP
Hdr
UDP
Hdr
Non-IKE
Mkr
ESP
Hdr
ESP
IV
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
20
20
12
20
2-257
12
Encrypted
This feature adds
16 bytes per packet
or 6400 bps per
G.729 call.
Authenticated
No additional overhead is caused by NAT transparency on a G.729 call over DSL (PPPoE/
AAL5) because there is enough AAL5 cell padding to absorb the 16 additional bytes per
packet (discussed in more detail in the following section). In cable implementations, these
additional 16 bytes increase the number of bits on the wire and, thus, the overall bandwidth
consumption. At the headend, NAT transparency increases the bandwidth consumption of
VoIP trafc by 1 Mbps for approximately every 82 concurrent G.729 calls; on the teleworker
router, NAT transparency increases the bandwidth required by 6.4 kbps per G.729 call.
Unless there is a need to implement this feature, the recommendation is to disable it when
bandwidth conservation is a requirement. This feature can be disabled with the no crypto
ipsec nat-transparency udp-encapsulation global conguration command.
DSL (AAL5 + PPPoE) Overhead

For DSL broadband connections, PPPoE is the most commonly implemented deployment.
Given the 112-byte, Layer 3 size of an IPSec (only) encrypted G.729 VoIP call, 40 bytes of
PPP, PPPoE, Ethernet, and ATM AAL5 headers and trailers are added. Additionally, the
pre-ATM Software Segmentation and Reassembly (SAR) engine is required to pad the
resulting 152-byte packet to the nearest multiple of 48 (each xed-length ATM cell can
transport only a 48-byte payload). Figure 16-24 shows the resulting 192-byte pre-ATM
packet.
676
Figure 16-24 IPSec-Encrypted G.729 Packet Size Through AAL5 + PPPoE Encapsulation
PPP
LLC 802.3 oE IPSec
Snap Hdr PPP
Hdr
10
14
ESP ESP
Hdr IV
20
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
20
12
20
12
AAL5 AAL5
Pad Trail
40
PPPoE + AAL5 Frame - 192 Bytes
The 192 bytes of cell payload are incorporated through ATM SAR into the (48-byte)
payloads of four 53-byte ATM cells (192 / 48 = 4 cells).
Therefore, the bandwidth required for an IPSec (only) encrypted G.729 VoIP call over DSL
is calculated as follows:
53 bytes per cell

4 cells per IPSec-encrypted VoIP packet
212 bytes per (192-byte) AAL5/PPPoE IPSec VoIP packet
10,600 bytes per second
8 bits per byte
Thus, an encrypted G.729 call requires 85 kbps of bandwidth, while an encrypted G.711
requires seven ATM cells or 148,400 bps on the wire.
This results in the LLQ conguration values of 85 kbps (priority 85) for a G.729 VoIP call
and 150 kbps for a G.711 (priority 150) VoIP call, respectively, for DSL.
Cable Overhead
For cable deployments, the Layer 2 overhead is less than that for DSL. The IPSec packet is
encapsulated in an Ethernet header (and trailer) that includes a 6-byte DOCSIS header, as
shown in Figure 16-25.
Figure 16-25 IPSec-Encrypted G.729 Packet Size Through Ethernet and DOCSIS Encapsulation
DOC 802.3 IPSec
SIS
Hdr
Hdr Hdr
14
20
ESP ESP
Hdr IV
8
IP
Hdr
UDP
RTP
Voice
ESP
Pad/NH
ESP
Auth
802.3
CRC
20
12
20
12
Ethernet + DOCSIS Frame - 136 Bytes
677
If baseline privacy is enabled (baseline privacy encrypts the payload between a cable
modem and the headend), the extended header is used, adding another 5 bytes.
The packet size of a G.729 call (with a zero-length extended header) is 136 bytes or
54,400 bps (at 50 pps); a G.711 call is 280 bytes or 112,000 bps (at 50 pps).
To simplify conguration and deployment, the values of 64 kbps (for G.729) and 128 kbps
(for either G.729 or G.711) can be used for priority queue denition for cable.
Asymmetric Links and Unidirectional QoS

With both DSL and cable, the uplink connection can be enabled with QoS, either in the
form of a service policy on the DSL (ATM PVC) interface or through a hierarchical MQC
service policy that shapes the uplink and prioritizes packets within the shaped rate on the
Ethernet interface. This half of the link is under the enterprises control and easily can be
congured.
The downlink connection is under the control of the broadband service provider, and any
QoS policy must be congured by the service provider; however, most service providers do
not offer QoS-enabled broadband services. This is usually because DSL providers often
have implemented non-Cisco equipment (DSLAM or other ATM concentration devices)
that typically have few or no QoS features available. Cable providers, on the other hand,
might have an option to enable QoS as DOCSIS 1.1 becomes more widely deployed.
Fortunately, most service offerings are asymmetrical. For example, consider a circuit with
a 256-kbps uplink and 1.5-Mbps downlink. The downlink rarely is congested to the point
of degrading voice quality.
Testing in the Cisco Enterprise Solutions Engineering labs has shown that when congestion
is experienced on the uplink, the resulting delay of (TCP) data packet acknowledgments
automatically decreases the arrival rate of downlink data trafc so that downlink congestion
does not occur. To summarize the results of these tests: Asymmetrical links are preferred
(over symmetrical links) as long as QoS is enabled on the uplink, provided that the lower
of the two speeds (the uplink) is adequate for transporting both voice and data.
Some service providers for business-class services offer symmetrical links in an effort to
compete with Frame Relay providers; 384 kbps/384 kbps and 768 kbps/768 kbps are
examples. With no QoS enabled on the service provider edge, this offering is not optimal
for the enterprise. An asymmetrical link such as 384 kbps/1.5 kbps is a better choice for the
V3PN networks.
678
Broadband Serialization Mitigation Through TCP Maximum Segment

Size Tuning
The majority of broadband deployments are DSL with PPPoE and cable with DOCSIS 1.0.
As previously noted, neither of these technologies includes any mechanisms to fragment
data packets and interleave voice packets at Layer 2 to minimize the impact of serialization
and blocking delay on voice packets (which is a recommended requirement on link speeds
768 kbps).
The DOCSIS 1.1 specication for cable includes fragmentation and interleaving support,
and DSL providers can implement MLP over ATM (which includes MLP LFI support).
However, these do not represent the majority of the currently deployed base of broadband
networks.
An alternative way to mitigate serialization delay on broadband circuits is provided by
adjusting the TCP Maximum Segment Size (MSS). The TCP MSS value inuences the
resulting size of TCP packets and can be adjusted with the ip tcp adjust-mss interface
command. Because the majority of large data packets on a network are TCP (normal UDP
application packets, such as DNS and NTP, average less than 300 bytes and do not create a
signicant serialization delay issue), this command effectively can reduce serialization
delay in most teleworker scenarios when no Layer 2 fragmentation and interleaving
mechanism is available.
NOTE
It is not recommended to run UDP-based video applications on broadband links 768 kbps
that do not support Layer 2 LFI in a V3PN teleworker scenario. This is because such
applications regularly generate large UDP packets that, obviously, are not subject to TCP
MSS and thus can cause signicant and unmitigatable serialization delays to VoIP packets.
The recommended TCP MSS value of 542 was calculated to eliminate the IPSec crypto
algorithm (3DES) padding and the ATM AAL5 padding in DSL implementations (cable
implementations have 3DES padding but no AAL5). Therefore, a TCP MSS value of 542
bytes is valid for cable but is optimized for DSL. Figure 16-26 illustrates a TCP packet with
an MSS size of 542.
NOTE
Both the ESP and AAL5 pad lengths are 0. A TCP packet with a TCP MSS value of 542 ts
exactly into 14 ATM cells, with no wasted bytes because of cell padding.
679
Figure 16-26 Optimized TCP MSS Value for DSL (542 Bytes)
AAL5
10
ATM
Hdr
IP
TCP
MISS
Size
20
20
542
IPSec
Hdr
ESP
SPI
ESP
Seq
ESP
IV
IP
TCP
MISS
Size
20
20
20
542
Ethernet
Header
PPPoE
SAR PDU
48 Bytes
ATM
Hdr
ESP
PAD Pad Len/
NH
0
ESP
Auth
12
Encrypted Packet
14
582
Bytes
632
SAR PDU
48 Bytes
ATM
Hdr
IPSec ESP
Tunnel Mode
SHA-1 3DES
632 Bytes
PAD
AAL5
672
Bytes
SAR PDU
48 Bytes
14 Cell Total
The evident negative aspect of using this technique to minimize the impact of serialization
delay is the decreased efciency of large data transfers, coupled with an increase in the
number of packets per second that the router must switch. The higher packets-per-second
rate is not as much of an issue at the remote router as at the headend, where hundreds or
thousands of remote connections are concentrated. However, adjusting the TCP MSS value
provides a means for the network manager to deploy voice over broadband connections,
which do not support any LFI mechanism at rates less than 768 kbps.
Split Tunneling
The teleworker is usually not the only user of the residential broadband connection. The
workers spouse and children also might utilize this link from other devices, such as
additional PCs or laptops or even gaming units. In such cases, only work-related trafc
would require encryption and could be given a preferential treatment, through QoS, over
general Internet trafc.
In this situation, prioritization could be made based on the destination address of the trafc.
One method to accomplish this would be to create a separate bandwidth class (for example,
named CORPORATE) and provision this class with a dedicated CBWFQ queue.
680
Given the sample topology illustrated in Figure 16-27, a suitable QoS conguration is
broken down as follows:
VOICEA single calls VoIP packets assigned to an LLQ, with the bandwidth
allocation based on codec and broadband media (DSL or cable).
CALL-SIGNALINGCall-Signaling trafc in a CBWFQ queue, allocated 5

percent.
INTERNETWORKInternetwork-Control trafc, such as routing protocol updates

and IKE keepalives, in a CBWFQ queue, allocated 5 percent.
CORPORATEAll other trafc that is destined to the enterprise through the IPSec
tunnel, in a CBWFQ queue, allocated 25 percent.
INTERNET/Class-DefaultTrafc to the Internet (outside the IPSec tunnel)

defaults to this fair-queued class.
Figure 16-27 Split Tunnel Example Topology

150.102.223.3
Spouse-and-Child PC
Cisco 837
DSL
V3PN
Teleworker
Router
Enterprise-Owned IP
Phone and Workstation
IP
Broadband/Internet
Service Provider
IPSec
Headend
Router(s)
150.102.223.4
The following conguration fragment provides a means to implement this policy. It assumes
that the headend IPSec crypto peers reside on network 150.102.223.0/29. In this example
environment, two peers are congured at 150.102.223.3 and 150.102.223.4. Packets within
the IPSec tunnelthose matching the CORP-INTRANET access control list (identifying
RFC 1918 private addresses that are assumed in this example to represent the intranets
behind the headends)will be encapsulated in an ESP (IPSec-IP protocol 50) IP header. A
class-map CORPORATE is congured that references the CORPORATE extended access
list, pointing to the VPN headend subnet.
A policy map named V3PN-SPLIT-TUNNEL is created that includes classes for VOICE,
INTERNETWORK-CONTROL, and CALL-SIGNALING, along with a bandwidth class
for CORPORATE.
681
In this example, the VOICE class is provisioned for a G.711 codec over a (384-kbps) DSL
uplink, allocating 150 kbps (or 40 percent, whichever syntax is preferred) for VoIP.
Example 16-8 shows the relevant conguration fragment.
Example 16-8 V3PN-SPLIT-TUNNEL Policy Example
!
crypto map SPLIT-TUNNEL-CRYPTO-MAP 1 ipsec-isakmp
set peer 150.102.223.3
set peer 150.102.223.4
set transform-set TS
! References CORP-INTRANET ACL
match address CORP-INTRANET
! Enables QoS Pre-Classify
qos pre-classify
!
!
match ip dscp ef
! VoIP
match ip dscp cs6
! IP Routing
match ip dscp cs3
match ip dscp af31
class-map match-all CORPORATE
! References CORPORATE ACL
match access-group name CORPORATE
!
policy-map V3PN-SPLIT-TUNNEL
class VOICE
priority 150
! Encrypted G.711 over DSL (PPPoE/AAL5)
bandwidth percent 5
bandwidth percent 5
class CORPORATE
! "Work-related" traffic provisioning
queue-limit 15
class class-default
fair-queue
queue-limit 15
!
...
!
ip access-list extended CORP-INTRANET
! CORP-INTRANET ACL (RFC 1918)
permit ip any 10.0.0.0 0.255.255.255
permit ip any 172.16.0.0 0.15.255.255
permit ip any 192.168.0.0 0.0.255.255
!
! ISAKMP ACL
!
ip access-list extended CORPORATE
! CORPORATE ACL (VPN Head-ends)
permit esp any 150.102.223.0 0.0.0.7
!
682
Teleworker V3PN QoS Designs

To review, three deployment models exist for teleworker V3PN scenarios:

Dual-Unit Model
Integrated Unit + Access Model
Furthermore, there are two main broadband deployment media: DSL and cable. DSL
supports all three teleworker deployment models, but cable supports (at the time of writing)
only the Integrated Unit + Access Model.
Integrated Unit/Dual-Unit ModelsDSL Design

The key point to remember when provisioning QoS for V3PN over DSL (PPPoE/ATM
AAL5) is the signicant bandwidth overhead required by these protocols (85 kbps is
needed per G.729 call, and 150 kbps is needed per G.711 call).
Some additional points to keep in mind are that since the service policy is being applied to
a low-speed ATM PVC, the Tx-ring should be tuned to 3 (as discussed in Chapter 13) and
also that because no LFI mechanism exists for DSL, TCP-MSS tuning can be done on the
dialer interface to mitigate serialization delay. An Integrated Unit/Dual-Unit V3PN
teleworker example for a 384-kbps DSL circuit is illustrated in Figure 16-28 and detailed
in Example 16-9.
Figure 16-28 Integrated Unit/Dual-Unit V3PN Teleworker Model for 384-kbps DSL Uplink
Best-Effort
50%
Voice - 150 kbps

40%
Call-Signaling 5%
683
Example 16-9 Integrated Unit/Dual-Unit V3PN Teleworker QoS Design Example for a 384-kbps (PPPoE/ATM)
DSL Uplink
!
class-map match-all
match ip dscp ef
class-map match-any
match ip dscp cs6
match access-group
class-map match-any
match ip dscp cs3
match ip dscp af31
VOICE
! VoIP
! IP Routing
name IKE
CALL-SIGNALING
!
!
policy-map V3PN-TELEWORKER
class VOICE
priority 150
bandwidth percent 5
bandwidth percent 5
class class-default
fair-queue
queue-limit 30


!
...
!
interface ATM0
no ip address
dsl operating-mode auto
dsl power-cutback 0
!
interface ATM0.35 point-to-point
description Outside PPPoE/ATM DSL Link
bandwidth 384
pvc dsl 0/35
vbr-nrt 384 384
tx-ring-limit 3
! Tx-Ring tuned to 3
pppoe max-sessions 5
service-policy output V3PN-TELEWORKER
! MQC policy applied to PVC
pppoe-client dial-pool-number 1
!
...
!
interface Dialer1
description Dialer for PPPoE
ip address negotiated
ip mtu 1492
encapsulation ppp
ip tcp adjust-mss 542
! TCP MSS value tuned for slow-link
!
continues
684
Example 16-9 Integrated Unit/Dual-Unit V3PN Teleworker QoS Design Example for a 384-kbps (PPPoE/ATM)
DSL Uplink (Continued)

...
!
!
! ISAKMP ACL
show policy
show atm pvc
Integrated Unit + Access ModelDSL/Cable Designs

Hierarchical MQC policies are required for Integrated Unit + Access Models to shape and
queue (within the shaped rate) on the outbound Ethernet interface. For cable, the shaped
rate should be 95 percent of the broadband links speed (as detailed in Chapter 13 in the
Frame Relay section [Committed Information Rate subsection]). For DSL, the shaped
rate should be 70 percent of the uplinks speed (to account for the increased bandwidth
overhead required by DSL). Furthermore, on slow-speed ( 768 kbps) links, TCP-MSS
should be tuned on both the inbound and outbound Ethernet interfaces. An Integrated Unit
+ Access Model for a 384-kbps cable teleworker uplink is shown in Figure 16-29 and
detailed in Example 16-10.
Figure 16-29 Integrated Unit + Access V3PN Teleworker Model for 384-kbps Cable Uplink
384 kbps
Cable Uplink Rate
Voice - 128 kbps

35%
Shaped to 95%
of Uplink Rate
Best-Effort
60%
Call-Signaling 5%
685
Example 16-10 Integrated Unit + Access ModelCable Design Example

!
class-map match-all
match ip dscp ef
class-map match-any
match ip dscp cs6
match access-group
class-map match-any
match ip dscp cs3
match ip dscp af31
VOICE
! VoIP
! IP Routing
name IKE
CALL-SIGNALING
!
!
policy-map V3PN-TELEWORKER
class VOICE
priority 128
bandwidth percent 5
bandwidth percent 5
class class-default
fair-queue
queue-limit 30
! Encrypted G.711 over Cable


!
!
policy-map SHAPE-384-CABLE
class class-default
! Shapes to 95% of 384 kbps cable link
service-policy V3PN-TELEWORKER
! Nested V3PN Teleworker queuing policy
!
...
!
interface Ethernet0
description Inside Ethernet Interface
!
interface Ethernet1
description Outside Ethernet Interface
ip address dhcp
! Shaper applied to LAN interface
service-policy output SHAPE-384-CABLE
!
show policy
686

ABC, Inc., is continuing to expand in its operations. Recently, it acquired a new subsidiary,
DEF Group, and wants to incorporate this new group within its existing network
infrastructure.
DEF already has provisioned its wide-area networking needs to its more than 100 branches
through IPSec VPNs. However, it has provisioned only for data networking needs. ABC,
Inc., wants to deploy IP Telephony to all of DEFs remote sites immediately. At this point
in time, DEFs branches are running full T1s, with Frame Relay as the access medium. ABC
also plans to deploy IP videoconferencing to these sites in the future, after the access links
have been upgraded to higher-speed ATM PVCs.
At the same time, ABC, Inc.s, management has received numerous requests from
employees who want to relocate away from the companys central campus because the cost
of living in other areas is more attractive. ABC is concerned that it will lose much of its top
talent and has decided to deploy a business-ready teleworker solution for any approved
employees.
This teleworker solution is to be tailored to work with either DSL or cable residential
broadband options through the Integrated Unit + Access Teleworker deployment model.
The combined set of V3PN solutions is illustrated in Figure 16-30.
In this example, ABC, Inc., is continuing to partner with a Cisco IP Multiservice Service
Provider (Service Provider XYZ) who offers them tight service-level agreements, such as
60-ms edge-to-edge delay, 10-ms edge-to-edge jitter, and 0.5 percent packet loss. However,
for this particular service (namely, the interconnect of DEFs remote sites), XYZ is not
offering explicit classes of service or QoS (beyond the edge-to-edge SLA).
For scalability and availability reasons, ABC, Inc., has decided to decouple IPSec VPN
encryption and QoS at its central site: one pair of routers (Cisco 7200 VXR NPE-G1s with
dual VPN Acceleration Modules, or VAMs) will provide IPSec VPN encryption, and
another pair of routers (Cisco 7500 with VIP6-80s) will perform WAN aggregation QoS.
As ABC continues to expand its operations, it plans to migrate its VPN headends to the
Catalyst 6500 with the VPN Services Module (VPNSM) to achieve even higher scalability.
Because ABC, Inc., plans to support IP videoconferencing in the futureand, therefore,
will be expanding the link speeds to its branchesit has decided to upgrade its branch
routers to Cisco 3725s and Cisco 3745s (both with AIM-II hardware crypto engines), with
the 3745s being deployed at the busiest sites.
For the telecommuter solution, ABC, Inc., has chosen a Cisco 831 router to perform the
basic, VPN/security, and QoS services required by teleworkers. It has decided to support
split-tunneling so that Internet trafc belonging to a spouse or child will not tax network
resources.
687
Figure 16-30 Case Study IPSec VPN QoS Design Example
Central
Campus
Tunnel 101
10.100.101.0/30
Branch
Tunnel 102
10.100.102.0/30
20.20.20.0/30
ATM OC3
VPN HE 1
WAG 1
VPN HE 2
WAG 2
Internet
Service
Provider
20.20.20.4/30
ATM OC3
Central Site
Intranet
192.168.1.0/24
Branch
Intranet
172.16.1.0/24
25.25.25.0/30
Frame Relay T1
Broadband
Service
Provider
VPN Headend
Subnet
100.100.100.0/24
ADSL
1.5 Mbps Downlink
384 kbps Uplink
DHCP
Teleworker
Teleworker
Intranet
10.200.1.0/24
The conguration solution for this example spans multiple routers: VPN headend routers
(see Example 16-11), WAN aggregators (see Example 16-12), branch routers (see Example
16-13), and telecommuter routers (see Example 16-14). To minimize redundancy, only a
single example of each router type is presented.
A key point to notice in the VPN headend designs, as detailed in Example 16-11, is that
QoS Pre-Classify is not required because encryption and QoS are being performed on
separate routers at the central site.
Example 16-11 Case Study Example: Part 1, VPN Headend Design
!
hostname VPN-HEADEND-1
!
ip cef
!
continues
688
Example 16-11 Case Study Example: Part 1, VPN Headend Design (Continued)
!
crypto isakmp policy 1
encr 3des
authentication pre-share
group 2
crypto isakmp key SECRETKEY address 25.25.25.1
!
!
crypto ipsec transform-set V3PN-TS esp-3des esp-sha-hmac
!
crypto map V3PN-CM 10 ipsec-isakmp
set peer 25.25.25.1
set transform-set V3PN-TS
match address V3PN-SITE1
[REPEAT CRYPTO MAP ENTRIES FOR EVERY REMOTE SITE, INCLUDING TELEWORKER SITES]
!
crypto map static-map local-address GigabitEthernet0/1
!
...
!
interface Tunnel101
ip address 10.100.101.1 255.255.255.252
! Applies crypto-map to tunnel interface
crypto map V3PN-CM
!
[REPEAT TUNNEL INTERFACES FOR EVERY REMOTE SITE, INCLUDING TELEWORKER SITES]
!
description CENTRAL SITE DATA SUBNET
ip address 192.168.1.1 255.255.255.0
duplex auto
speed auto
media-type rj45
no negotiation auto
!
!
description VPN HEADEND SUBNET
ip address 100.100.100.1 255.255.255.0
duplex auto
speed auto
media-type rj45
no negotiation auto
! Applies crypto map to GE 0/2
crypto map V3PN-CM
!
...
!
ip access-list extended V3PN-SITE1
permit gre host 100.100.100.1 host 25.25.25.1
!
689
show policy
Some key points to notice in the WAN aggregator designs, as detailed in Example 16-12,
include these:
IP CEF is required to be run in distributed mode on this platform (Cisco 7500).
Because the central sites link to the ISP is a (very) high-speed ATM PVC (OC3), no
tuning of the Tx-ring is required.
ISAKMP trafc is not encrypted and thus can be classied by a simple ACL (UDP
port 500).
A six-class V3PN policy has been deployed on the WAN aggregators edge.
WRED in conjunction with queue-limit tuning is supported on distributed platforms
(such as the Cisco 7500), although it currently is not supported on nondistributed
platforms (until the release of Consistent QoS Behavior, as described in Chapter 13).
Example 16-12 Case Study Example: Part 2, WAN Aggregator QoS Design
!
hostname WAG-1
!
ip cef distributed
!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
match ip dscp cs6
match ip dscp cs1
!
!
class VOICE
priority percent 33
bandwidth percent 5
bandwidth percent 5
! Distributed platform
! VoIP
! IP Routing
! Scavenger
! VoIP gets 33% LLQ

continues
690
Example 16-12 Case Study Example: Part 2, WAN Aggregator QoS Design (Continued)
queue-limit 20
class SCAVENGER
bandwidth percent 1
queue-limit 1
class class-default
queue-limit 16
!
...
!
interface GigabitEthernet1/0/0
description VPN HEADEND SUBNET
ip address 100.100.100.3 255.255.255.0
duplex auto
speed auto
media-type rj45
no negotiation auto
!
...
!
interface ATM1/1/0
no ip address
!
interface ATM1/1/0.100 point-to-point
description ATM OC3 TO ISP
ip address 20.20.20.1 255.255.255.252
pvc 0/100
vbr-nrt 149760 149760
service-policy out SIX-CLASS-V3PN-EDGE
!
!
...
!
!
show policy
! WRED + QL tuning supported on 7500
! WRED + QL tuning supported on 7500
! MQC policy applied to PVC
! ISAKMP ACL
691
Some key points to notice in the V3PN branch router designs, as detailed in Example 16-13,
include these:
A six-class V3PN policy has been deployed on the branch VPN access edge.
Even though Frame Relay is being used as the access medium, no shaping is required
because all sites are running at full (T1) port speeds.
QoS Pre-Classify is enabled on both crypto map entries and tunnel interfaces, to
enhance performance.
The delay parameter has been tuned on the tunnel interfaces to inuence the routing
protocols (EIGRP, in this instance) choice for primary and secondary tunnels.
WRED is not supported in conjunction with queue-limit tuning (for Anti-Replay) on

these nondistributed platforms.
Example 16-13 Case Study Example: Part 3, V3PN Branch Router Design
!
hostname V3PN-BRANCH-1
!
ip cef
!
!
encr 3des
authentication pre-share
group 2
!
!
crypto ipsec transform-set V3PN-TS esp-3des esp-sha-hmac
!
set peer 100.100.100.1
match address V3PN-HE1
qos pre-classify
set peer 100.100.100.2
match address V3PN-HE2
qos pre-classify
!
crypto map static-map local-address Serial0/0
!
continues
692
Example 16-13 Case Study Example: Part 3, V3PN Branch Router Design (Continued)
!
match ip dscp ef
match ip dscp cs3
match ip dscp af31
match ip dscp cs6
match ip dscp cs1
!
!
class VOICE
priority percent 33
bandwidth percent 5
bandwidth percent 5
queue-limit 20
class SCAVENGER
bandwidth percent 1
queue-limit 1
class class-default
queue-limit 16
!
!
...
!
interface Tunnel101
ip address 10.100.101.2 255.255.255.252
delay 50000
qos pre-classify
crypto map V3PN-CM
!
interface Tunnel102
ip address 10.100.102.2 255.255.255.252
delay 60000
qos pre-classify
crypto map V3PN-CM
!
!
! VoIP
! IP Routing
! Scavenger
! VoIP gets 33% LLQ

693
Example 16-13 Case Study Example: Part 3, V3PN Branch Router Design (Continued)
description V3PN-BRANCH-1 INTRANETS
ip address 172.16.1.1 255.255.255.0
duplex auto
speed auto
!
...
!
interface Serial0/0
description FRAME RELAY T1 TO ISP
bandwidth 1536
ip address 25.25.25.1 255.255.255.252
service-policy output SIX-CLASS-V3PN-EDGE ! MQC policy applied to T1 int
crypto map V3PN-CM
!
...
!
! ISAKMP ACL
!
ip access-list extended V3PN-HE1
!
ip access-list extended V3PN-HE2
!
show policy
Some key points to notice in the V3PN teleworker router designs, as detailed in Example 16-14,
include these:
In this example, an ADSL link (of 1536 kbps downlink and 384 kbps uplink)
is provisioned for V3PN. Because of the overhead of DSL, VoIP bandwidth is
provisioned for 150 kbps (which can support either G.711 or G.729).
A hierarchical QoS policy is set to shape (and queue within the shaped rate) to
70 percent of the uplinks rate and is applied to the Ethernet interface facing the
DSL modem.
WRED is not supported in conjunction with queue-limit tuning (for Anti-Replay) on

these nondistributed platforms.
NAT transparency is disabled globally to reduce bandwidth overhead.
694
QoS Pre-Classify is enabled on the crypto map to enhance performance.
TCP MSS is adjusted on both Ethernet interfaces to mitigate serialization delay.
A split-tunneling policy is provisioned on teleworker routers, determining workrelated trafc as being destined to private intranet addresses.
Example 16-14 Case Study Example: Part 4, Telecommuter Router

!
hostname TELEWORKER-V3PN-1
!
ip cef
ip inspect name CBAC tcp
ip inspect name CBAC udp
ip inspect name CBAC ftp
!
encr 3des
group 2
!
!
crypto ipsec transform-set V3PN-TELEWORKER-TS esp-3des esp-sha-hmac
no crypto ipsec nat-transparency udp-encaps ! Disables NAT Transparency
!
crypto map V3PN-TELEWORKER-CM 1 ipsec-isakmp
set peer 100.100.100.1
set peer 100.100.100.2
set transform-set V3PN-TELEWORKER-TS
match address CORP-INTRANET
qos pre-classify
!
!
! VoIP
match ip dscp ef
! IP Routing
match ip dscp cs6
match ip dscp cs3
match ip dscp af31
class-map match-all CORPORATE
! References CORPORATE ACL
match access-group name CORPORATE
!
!
policy-map V3PN-SPLIT-TUNNEL
class VOICE
priority 150
bandwidth percent 5
bandwidth percent 5
695
Example 16-14 Case Study Example: Part 4, Telecommuter Router (Continued)

class CORPORATE
queue-limit 15
class class-default
fair-queue
queue-limit 15
! "Work-related" traffic provisioning


!
!
policy-map SHAPE-384-ADSL-UPLINK
class class-default
! Shapes to 70% of 384 kbps ADSL uplink
service-policy V3PN-SPLIT-TUNNEL
! Nested V3PN split-tunnel policy
!
...
!
interface Ethernet0
description TELEWORKER 1 INTRANET
ip address 10.200.1.1 255.255.255.248
ip inspect CBAC in
!
interface Ethernet1
description ETHERNET LINK TO DSL MODEM
ip address dhcp
! Shaping + Queuing
service-policy output SHAPE-384-ADSL-UPLINK
duplex auto
no cdp enable
crypto map V3PN-TELEWORKER-CM
!
!
! CORP-INTRANET ACL (RFC 1918)
ip access-list extended CORP-INTRANET
permit ip 10.200.1.0 0.0.0.7 10.0.0.0 0.255.255.255
permit ip 10.200.1.0 0.0.0.7 172.16.0.0 0.15.255.255
permit ip 10.200.1.0 0.0.0.7 192.168.0.0 0.0.255.255
!
! ISAKMP ACL
!
ip access-list extended CORPORATE
! CORPORATE ACL (VPN Head-ends)
permit esp any 100.100.100.0 0.0.0.3
!
show policy
696
Summary
IPSec VPNs, the most commonly deployed VPN solutions today, are found in three main
contexts: site-to-site VPNs, teleworker VPNs, and remote-access VPNs. The overlaying of
QoS technologies on top of IPSec VPNs is dubbed V3PN, for voice- and video-enabled
Virtual Private Networks. This chapter presented considerations and design recommendations for V3PN deployments in site-to-site and teleworker contexts. A summary of the
design recommendations for encryption and QoS for site-to-site and teleworker IPSec
V3PNs is illustrated in Figure 16-31.
Figure 16-31 IPSec V3PN Site-to-Site and Teleworker Design Summary Comparison
Encryption
IPSec/GRE
LLQ/CBWFQ/LFI
HDLC
MLP
ATM
WAN Media
Site-to-Site
QoS
Frame
Relay
Ethernet
IPSec Only
Hierarchical Shaping +
Queuing + TCP MSS
DSL
Cable
Broadband Media
Teleworker
Some site-to-site considerations that were discussed include the bandwidth implications
of various IPSec modes of operations and the incompatibility of cRTP and IPSec. The
interrelation of IPSec VPN logical hub-and-spoke topologies and the effect these have on
spoke-to-spoke delay budgets also were examined. Subsequently, the ToS byte preservation
mechanism was overviewed along with the QoS Pre-Classify Cisco IOS Software feature,
which allows for the classication of packets already encrypted (on the same router)
through ACLs. QoS and Anti-Replay implications then were discussed, illustrating how
QoS policies that reorder packets potentially can exacerbate Anti-Replay drops (an IPSec
message-integrity mechanism). Techniques for minimizing such undesired QoS/Anti-Reply
interaction effects, such as reducing the queue length of data queues, were presented. Next,
the need for control plane provisioning was highlighted, along with basic designs for doing so.
Several site-to-site QoS models were detailed, ranging from a six-class V3PN QoS model
to a complex 11-class V3PN QoS Baseline model. WAN aggregator considerations specic
to IPSec VPN deployments were examined next, including QoS provisioning for IPSec
over private WANs, per-tunnel hierarchical shaping and queuing, and recommendations for
decoupled VPN headend/WAN aggregation deployment models, where encryption and
QoS are performed on different routers.
Further Reading
697
Attention then shifted to teleworker scenarios and the three main teleworker deployment
models: the Integrated Unit Model, the Dual-Unit Model, and the Integrated Unit + Access
Model. The two main broadband media types, DSL and cable, were broken down to
ascertain the bandwidth-provisioning implications of each media. Neither DSL nor
(DOCSIS 1.0) cable includes any mechanism for serialization delay mitigation, so TCP
maximum segment-size tuning was considered as an alternative mechanism to achieve this.
Split-tunneling designs, to address spouse-and-child requirements, were introduced.
Teleworker V3PN designs then were detailed for Integrated Unit and Dual-Unit models
over DSL, in addition to an Integrated Unit + Access Model solution for cable.
The chapter concluded with a case study (which, likewise, concluded the case studies
presented in each previous design chapter) highlighting how an enterprise could overlay
QoS on top of an existing site-to-site IPSec VPN to enable a V3PN. Additionally, a solution
was presented for a business-ready teleworker.
Further Reading
Standards:
RFC 1321, The MD5 Message-Digest Algorithm: http://www.ietf.org/rfc/

rfc1321.txt.
RFC 1918, Address Allocation for Private Internets: http://www.ietf.org/rfc/

rfc1918.txt.
RFC 2104, HMAC: Keyed-Hashing for Message Authentication:

RFC 2401, Security Architecture for the Internet Protocol: http://www.ietf.org/

rfc/rfc2401.txt.
RFC 2402, IP Authentication Header: http://www.ietf.org/rfc/rfc2402.txt.
RFC 2404, The Use of HMAC-SHA-1-96 within ESP and AH: http://www.ietf.org/
rfc/rfc2404.txt.
RFC 2405, The ESP DES-CBC Cipher Algorithm with Explicit IV:
RFC 2406, IP Encapsulating Security Payload (ESP): http://www.ietf.org/rfc/

rfc2406.txt.
RFC 2407, The Internet IP Security Domain of Interpretation for ISAKMP:

RFC 2408, Internet Security Association and Key Management Protocol

(ISAKMP): http://www.ietf.org/rfc/rfc2408.txt.
RFC 2403, The Use of HMAC-MD5-96 Within ESP and AH: http://www.ietf.org/
rfc/rfc2403.txt.
698
RFC 2409, The Internet Key Exchange (IKE): http://www.ietf.org/rfc/rfc2409.txt.
RFC 2411, IP Security Document Roadmap: http://www.ietf.org/rfc/rfc2411.txt.
RFC 2410, The NULL Encryption Algorithm and Its Use with IPSec:
RFC 2412, The OAKLEY Key Determination Protocol: http://www.ietf.org/rfc/
rfc2412.txt.
Books:
Kaeo, Merike. Designing Network Security. Indianapolis: Cisco Press, 2003.
Mason, Andrew. Cisco Secure Virtual Private Networks. Indianapolis: Cisco Press,
2001.
Malik, Saadat. Network Security Principles and Practices. Indianapolis: Cisco Press,
2002.
Cisco IOS documentation:
IP Security and Encryption overview (Cisco IOS Release 12.2):

fsecur_c/psenc/scfencov.htm.
Conguring IPSec network security (Cisco IOS Release 12.2): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122cgcr/fsecur_c/psenc/scpsec.htm.
Conguring Internet Key Exchange Security Protocol (Cisco IOS Release 12.2):
fsecur_c/psenc/scke.htm.
Prefragmentation for IPSec VPNs (Cisco IOS Release 12.2[13]T):

http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/
122newft/122t/122t13/ftprefrg.htm.
Quality of service for Virtual Private Networks (Cisco IOS Release 12.2[2]T):
122t/122t2/ftqosvpn.htm.
Low-latency queuing (LLQ) for IPSec encryption engines (Cisco IOS Release
12.2[13]T): http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/
122newft/122t/122t13/llqfm.htm.
IPSec NAT transparency (Cisco IOS Release 12.2[13]T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122newft/122t/122t13/ftipsnat.htm.
IPSec and quality of service feature (Cisco IOS Release 12.3[8]T):

123t/123t_8/gtqosips.htm.
Further Reading
Conguring broadband access (Cisco IOS Release 12.2): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122cgcr/fwan_c/wcfppp.htm.
PPP over Ethernet Client (Cisco IOS Release 12.2[2]T): http://www.cisco.com/

univercd/cc/td/doc/product/software/ios122/122newft/122t/122t2/ftpppoec.htm.
699
QoS At-A-Glance Summaries
QoS Tools
QoS is the measure of transmission quality and service
availability of a network (or internetworks). The
transmission quality of the network is determined
by the following factors: Latency, Jitter, and Loss.
Classification can be done at Layers 2-7

via the following tools:
Layer 2MAC access lists
Layer 3/4IP access lists
Layer7Network-Based Application Recognition
L2 Frame
L3 IP Packet
Policing tools can complement marking tools by

marking metering flows and marking-down
out-of-contract traffic.
L4 Segment
L7 Data Payload
NBAR PDLM
Delay
(Latency)
DelayVariation
(Jitter)
Packet
Loss
QoS technologies refer to the set of tools and

techniques to manage network resources and are
considered the key enabling technologies for the
transparent convergence of voice, video, and data
networks. Additionally, QoS tools can play a
strategic role in significantly mitigating
DoS/worm attacks.
The Cisco QoS toolset consists of the following:
Classification and Marking tools
Policing and Markdown tools
Scheduling tools
Link-specific tools
AutoQoS tools
Classification
and Marking
STOP
Policers meter traffic into three categories:
Marking can be done at Layers 2 or Layer 3:

Layer 2: 802.1Q/p CoS, MPLS EXP
Layer 3: IP Precedence, DSCP, and/or
IP ECN
Conform: Traffic is within the defined

rate (green light)
Exceed: Moderate bursting is allowed
(yellow light)
Violate: No more traffic is allowed
beyond this upper-limit (red light)
used
Layer 3 (IP ToS Byte) Marking Options:

7
IP Precedence

RFC 2474
DiffServ Extensions
Un
Scheduling tools reorder and selectively-drop

packets whenever congestion occurs.
IP ECN
RFC 3168
IP ECN Bits
Cisco recommends end-to-end marking at Layer 3

with standards-based DSCP values.
Policing and
Markdown
Link-specific tools are useful on slow-speed

WAN/VPN links and include shaping, compression,
fragmentation, and interleaving.
AutoQoS features automatically configure
Cisco-recommended QoS on Catalyst switches and
IOS routers with just one or two commands.
Traffic Shaping
Scheduling
(Queuing and
Selective-Dropping)
Link-Specific
Mechanisms
cut here
cut here
The Cisco QoS Baseline

The QoS Baseline is a strategic
document designed to unify QoS
within Cisco. The QoS Baseline
provides uniform, standards-based
recommendations to help ensure
that QoS products, designs, and
deployments are unified and
consistent.
The QoS Baseline defines up to
11 classes of traffic that may be
viewed as critical to a given
enterprise. A summary these
classes and their respective
standards-based markings
and recommended QoS
configurations are shown
below.
Application
1
IP Routing
L3 Classification
PHB
DSCP
CS6
48
Rate-Based Queuing + RED
Voice
EF
46
RFC 3246
RSVP Admission Control + Priority Queuing
Interactive-Video
AF41
34
RFC 2597
RSVP + Rate-Based Queuing + DSCP-WRED
Streaming-Video
CS4
32
RFC 2474-4.2.2
RSVP + Rate-Based Queuing + RED
AF31
26
RFC 2597
Rate-Based Queuing + DSCP-WRED
Call-Signaling
CS3
24
RFC 2474-4.2.2
Transactional Data
AF21
28
RFC 2597
CS2
16
RFC 2474-4.2.2

8 Network-Management
9
Bulk Data
AF11
Bulk Data
10
RFC 2597
10
Scavenger
CS1
Scavengor
Internet 2
No BW Guarentee + RED
11
Best-Effort
0Best Effort
RFC 2474-4.1
BW Guarentee Rate-Based Queuing + RED
The IP Routing class is intended for IP Routing protocols, such as BGP, OSPF, etc.
Voice refers to VoIP bearer traffic only (and does not include Call-Signaling traffic).
Interactive-Video refers to IP Video-Conferencing.
Streaming-Video is either unicast or multicast unidirectional video.
The (Locally-Defined) Mission-Critical Data class is intended for a subset

of Transactional Data applications that contribute most significantly to the
business objectives (this is a non-technical assessment).
The Call-Signaling class is intended for voice and/or video signaling traffic,
such as Skinny, SIP, H.323, etc.
RFC 2474-4.2.2
Recommended Configuration
13
Referencing
Standard 12
The Transactional Data class is intended for foreground, user-interactive applications,

such as database access, transaction services, interactive messaging, and preferred
data services.
The Network-Management class is intended for network management protocols,
such as SNMP, Syslog, DNS, etc.
Cisco products that

support QoS features
use these QoS Baseline
recommendations
for marking and
scheduling and
admission control.
The QoS Baseline
recommendations are
intended as a standardsbased guideline for
customersnot as
a mandate.
The Bulk Data class is intended for background, non-interactive traffic

flows, such as large file transfers, content distribution, database
synchronization, backup operations, and e-mail.
10
The Scavenger class is based on an Internet 2 draft that defines a less-than

best-effort service. In the event of link congestion, this class will be
dropped the most aggressively.
11
The Best-Effort class is also the default class. Unless an application has been
assigned for preferential/deferential service, it remains in this default class.
Most enterprises have hundredsif not thousandsof applications on their
networks; the majority of which will remain in the Best-Effort service class.
12
Standards-based marking recommendations allow for better integration with

service-provider offerings as well as other internetworking scenarios.
13
In Cisco IOS, rate-based queuing translates to CBWFQ; priority queuing is

LLQ.DSCP-Based WRED (based on RFC 2597) drops AFx3 before AFx2 and
in turn drops AFx2 before AFx1. RSVP is recommended (whenever supported)
for Voice and/or Interactive-Video admission control
QoS Best-Practices
A successful QoS deployment includes three
key phases:
Strategically defining the business objectives to
be achieved via QoS.
Analyzing the service-level requirements of
the traffic classes.
Designing and testing QoS policies
1 Strategically defining the business objectives to
be achieved by QoS.
Business QoS objectives need to be defined:
Is the objective to enable VoIP only or is video
also required?
If so, is video-conferencing required or streaming
video? Or both?
Are there applications that are considered
mission-critical? If so, what are they?
Does the organization wish to squelch certain
types of traffic? If so, what are they?
Does the business want to use QoS tools to
mitigate DoS/worm attacks?
How many classes of service are needed to
meet the business objectives?
Because QoS introduces a system of managed unfairness,
most QoS deployments inevitably entail political and
organizational repercussions when implemented. This is
typically because disagreement often arises over which
applications are to be considered "mission-critical,"
which is a non-technical, subjective evaluation (MissionCritical applications are foreground, interactive
applications that most directly contribute to the main
business objective of the enterprise).
To minimize the effects of these non-technical
obstacles to deployment, address these political
and organizational issues as early as possible,
garnering executive endorsement whenever
possible.
2 Analyzing the application service-level requirements.
3 Designing and testing the QoS policies.

Classify, mark, and police as close to the trafficsources as possible; following Differentiated
Services standards, such as RFCs 2474, 2475,
2597, 2698, and 3246.
L3 Classification
Application
Predictable Flows
Drop + Delay Sensitive
UDP Priority
150-ms one-way delay
30-ms jitter
1% loss
17 kbps-106 kbps VoIP
+Call-Signaling
Voice
Unpredictable Flows
Drop + Delay Sensitive
Video
UDP Priority
150-ms one-way delay
30-ms jitter
1% loss
Overprovision stream by 20% to
account for headers + bursts
PBH
DSCP
Routing
CS6
48
Voice
EF
46
Interactive-Video
AF41
34
32
Streaming-Video
CS4
AF31
26
Call-Signaling
CS3
24
Transactional Data
AF21
28
Network-Management
CS2
16
10
Bulk Data
AF11
Bulk Data
Scavengor
CS1
Scavengor
Best-Effort
0Best Effort
Provision queuing in a consistent manner (according

to platform capabilities). For example, in the campus
(inner circle) there may be only four hardware queues
available; however, on the WAN/VPN (outer circle),
eleven software queues may be used. The point is to
keep the respective bandwidth allocations consistent.
Voice
Best-Effort
Scavenger
Best-Effort
25%
Real-Time
33%
InteractiveVideo
Bulk Data
Critical Data
No "one-size fits all"

Smooth/Bursty
Benign/Greedy
TCP retransmits/
UDP does not
Data
Streaming-Video
Internetwork
Control
Network-Management
Transactional Data Mission-Critical Data
Thoroughly test QoS policies prior to

production-network deployment.
cut here
Call-Signaling
cut here
A key point of this strategy is that legitimate traffic

flows which temporarily exceed thresholds are not
penalized by Scavenger-class QoS.
Scavenger-Class QoS Strategy

for DoS/Worm Attack Mitigation
DoS and worm attacks are exponentially increasing
in frequency, complexity, and scope of damage.
QoS tools and strategic designs can mitigate the
effects of worms and keep critical applications
available during DoS attacks.
Only traffic in excess of the normal/abnormal

threshold is re-marked to Scavenger.
The Campus uplinks are not the only points in

the network infrastructure that congestion can
occur. Typically WAN and VPN links are the first
to congest.
Scavenger/DSCP CS1
Normal/Abnormal Threshold
One such strategy, referred to as Scavenger-class

QoS, uses a two-step tactical approach to provide
first- and second-order anomaly detection and
reaction to DoS/worm attack-generated traffic.
The first step in deploying Scavenger-class QoS is
to profile applications to determine what constitutes
a normal versus abnormal flow (within a 95%
confidence interval).
Only sustained, abnormal streams generated

simultaneously by multiple hosts (highly-indicative
of DoS/worm attacks) are subject to aggressive
droppingand such dropping only occurs after
legitimate traffic has been fully-serviced.
Therefore, Scavenger-class less-than best-effort

queuing should be provisioned on all network
devices in a consistent manner (according to
platform capabilities).
Voice
Campus Access-Edge policing policies are coupled

with Scavenger-class queuing policies on the uplinks
to the Campus Distribution Layer.
Best-Effort
Application traffic exceeding this normal rate is

subject to first-order anomaly detection at the
Campus Access-Edge: specifically, excess traffic
is marked down to Scavenger (DSCP CS1/8).
Queuing policies only engage when links are

congested. Therefore, only if uplinks become
congested does traffic begin to be dropped.
Note that anomalous traffic is not dropped or

penalized at the edge; it is simply re-marked.
Anomalous trafficpreviously marked to Scavenger

is dropped the most aggressively (only after all other
traffic types have been fully-serviced).
Policing Policy
Policing Policy
Normal Traffic
Normal Traffic
Scavenger
Best-Effort
25%
Bulk Data
Real-Time
35%
InteractiveVideo
Critical
20%
StreamingVideo
Call-Signaling
NetworkManagement
Routing
Transactional
Data
Mission-Critical
Data
Thoroughly test QoS policies prior to

production-network deployment.
Anomalous Traffic
Anomalous Traffic
Queuing Policy
QoS policies should always be enabled in Catalyst

switch hardwarerather than router software
whenever a choice exists.
Three main types of QoS policies are required within
the campus:
Classification and Marking
Policing and Markdown
Queuing
Classification, marking, and policing should be
performed as close to the traffic sources as possible,
specifically at the Campus Access-Edge. Queuing,
on the other hand, needs to be provisioned at all
campus layers (access, distribution, core) due to
oversubscription ratios.
Classify and mark as close to the traffic sources as
possible following the Cisco QoS Baseline marking
recommendations, which are based on Differentiated
Services standards, such as: RFC, 2474, 2597, and 3246.
Application
Routing
Campus QoS Design

Access-Edge policers, such as this one, detect
anomalous flows and re-mark these to Scavenger
(DSCP CS1).
DSCP
CS6
48
VVLAN +
DSCP EF
Yes
EF
46
Interactive-Video
AF41
34
CoS
CoS 7
CS6
CoS 6
EF
CoS 5
AF41
CoS 4
CS4
CoS 4
DSCP 25
CoS 3
AF31/CS3
CoS 3
AF21
CoS 2
CS2
CoS 2
AF11
CoS 1
1P3Q1T
No
Yes
Trust and Transmit
No
CoS 6
Drop
VVLAN +
DSCP
CS3
Yes
32 kbps
Yes
32 kbps
DVLAN
ANY
Yes
Re-Mark to DSCP CS1
5 Mbps
Yes
Re-Mark to DSCP 0
Streaming-Video
CS4
32
Mission-Critical
AF31
26
Call-Signaling
CS3
24
Transactional Data
AF21
28
Network-Management
CS2
16
Bulk Data
AF11
Bulk Data
10
Trust-DSCP + Queuing
Conditional Trust +
Policing + Queuing
Scavengor
CS1
Scavengor
Untrusted + Policing +
Queuing
Best-Effort
0Best Effort
Per-User Microflow
Policing
No
CoS 3
CoS 0
Re-Mark to DSCP 0
No
Yes
Queue 3
(70%)
CoS 2
Re-Mark to DSCP CS1
Yes
CoS 4
Trust and Transmit
No
No
VVLAN
ANY
Q4
Priority Queue
CoS 7
128 kbps
No
Voice
DSCP
CoS 5
Start
L3 Classification
PBH
Queuing policies will vary by platform:

e.g., 1P3Q1T P = Priority Queue
Q = Non-Priority Queue
T = WRED Threshold
CS1
CoS 1
Queue 2
(25%)
Queue 1 (5%)
CoS 1
Queue 1 (5%)
The diagram below shows what QoS policies are

needed where in the Campus.
Re-Mark to DSCP CS1
VVLAN = Voice VLAN

DVLAN = Data VLAN
Server Farms
cut here
IP Phones + PCs
IP Phones + PCs
cut here
In an enterprise network infrastructure, bandwidth

is scarcestand thus most expensiveover the WAN.
Therefore, the business case for efficient bandwidth
optimization via QoS technologies is strongest over
the WAN.
WAN QoS Design

Queuing model examples for 5/8/11 classes of service
are shown below:
WAN QoS policies need to be configured on the WAN

edges of WAN Aggregator (WAG) routers and Branch
routers. WAN-edge QoS policies include queuing,
shaping, selective dropping, and link-specific policies.
The number of WAN classes of traffic is determined
by the business objectives and may be expanded
over time.
5-Class Model
Real-Time
8-Class Model
Voice 18%
Call-Signaling
Call-Signaling
Branch
Use MLP LFI and cRTP on 768-kbps links

Frame-Relay Link
Scavenger 1%
Voice
InteractiveVideo 15%
Bulk Data
4%
WAG
StreamingVideo 10%
Interactive-Video
Video
Leased-Line (MLP) Link
WAG
Best-Effort
25%
QoS Baseline
Model
Voice
Link-Specific Design Recommendations:
Transactional
Data 7%
Streaming-Video
Call-Signalling
Mission-Critical
Data 10%
Call-Signaling
5%
Routing 3%
Network-Management
Frame Relay
Cloud
Branch
Use Frame-Relay traffic shaping

Set CIR to 95% of guaranteed rate
Set Committed Burst to CIR/100
Set Excess Burst to 0
Use FRF.12 and and cRTP on 768-kbps links
ATM Link
IP Routing
Network Control
WAN QoS tools: RTP Header Compression (cRTP)

Network-Management
IP Header
20 Bytes
Critical Data
UDP Hdr
8 Bytes
RTP Hdr
12 Bytes
VoIP
Transactional Data
Bulk Data
cRTP Header
2-5 Bytes
Best-Effort
Best-Effort
Scavenger
Scavenger
Scavenger
ATM
Cloud
Branch
Use MLP LFI (via MLPoATM) and cRTP on

slow-speed links
Set the ATM PVC Tx-ring to 3 for 768-kbps links
cRTP saves:
~20% for G.711
~60% for G.729
Bulk Data
Best-Effort
WAG
ATM-to-FR SIW Link

ATM
Cloud
WAN QoS tools: Link Fragmentation and Interleaving

WAG
Data
VoIP
FR
Cloud
Branch
Time
WAN links can be categorized into three main

speed groups:
Slow-Speed ( 768 kbps)
Medium-Speed (> 768 kbps & T1/E1)
High-Speed ( T1/E1)
LFI tools (MLP LFI or FRF.12) fragment large data

packets and interleave these with high-priority VoIP.
Data
Data
Data
VoIP
Data
Use MLP LFI (via MLPoATM and MLPoFR) for

768-kbps links
Optimize fragment sizes to minimize ATM
cell-padding
Branch QoS Design

Branch routers are connected to central sites via
private-WAN or VPN links, which often prove to be
the bottlenecks for traffic flows. QoS policies at these
bottlenecks align expensive WAN/VPN bandwidth
utilization with business objectives.
QoS designs for Branch routers arefor the most
partidentical to WAN Aggregator QoS designs.
However, Branch routers require three unique
QoS considerations:
1) Unidirectional applications
2) Ingress classification requirements
3) NBAR policies for worm policing
Each of these Branch router QoS design
considerations is reviewed.
3 NBAR for Known Worm Policing
An example 10-Class QoS Baseline Branch Router

WAN Edge Queuing Model:
Voice 18%
Best-Effort
25%
1. The enabling code
Scavenger 1%
Bulk Data
4%
2. The propagation
mechanism
3. The payload
Call-Signaling
5%
Routing 3%
Network-Management
2%
Misson-Critical
Data 15%
Transactional
Data 12%
1 Unidirectional Applications
L2 Frame
Some applications (like Streaming-Video) usually

only traverse the WAN/VPN in the campus-to-branch
direction and, therefore, do not require provisioning in
the branch-to-campus direction on the branch
routers WAN edge.
The branch routers ingress LAN edge is a strategic

place to use NBAR to identify and drop worms, such
as CodeRed, NIMDA, SQL Slammer, MS-Blaster,
and Sasser.
L3 IP Packet
L4 Segment
L7 Data Payload
Worm
2 Ingress Classification
Bandwidth for such unidirectional application classes

can be reassigned to other critical classes, as shown
in the following diagram. Notice that no StreamingVideo class is provisioned and the bandwidth allocated
to it (on the campus side of the WAN link) is
reallocated to the Mission-Critical and Transactional
Data classes.
Worms are nothing new, but they have increased

exponentially in frequency, complexity and scope
of damage in recent years.
Branch-to-campus traffic may not be correctly

marked on the branch access-layer switch.
These switcheswhich are usually lower-end
switchesmay or may not have the capabilities to
classify and mark application traffic. Therefore,
classification and marking may need to be performed
on the branch routers LAN edge
(in the ingress direction).
NBAR extensions allow for custom Packet Data

Language Modules (PDLMs) to be defined for
future worms.
Where is QoS required on Branch routers?
LLQ/CBWFQ/WRED/Shaping/
LFI/cRTP Policies for
Classification and Marking +

NBAR Worm Policing Policies
Branch Router
Branch Switch
DVLAN
WAN/VPN
VVLAN
Furthermore, branch routers offer the ability to use

NBAR to classify and mark traffic flows that require
stateful packet inspection.
cut here
WAN Edge
LAN Edge
Optional: DSCP-to-CoS Mapping Policies
for Campus-to-Branch Traffic
cut here
QoS design for an enterprise subscribing to an MPLS

VPN requires a major paradigm shift from privateWAN QoS design.
This is because with private-WAN design, the enterprise
principally controlled QoS. The WAN Aggregator
(WAG) provisioned QoS for not only campus-tobranch traffic, but also for branch-to-branch traffic,
which was hmed through the WAG.
Branch
WAG
WAN
Branch
However, due to the any-to-any/full-mesh nature of

MPLS VPNs, branch-to-branch traffic is no longer
homed through the WAG. While branch-to-MPLS
VPN QoS is controlled by the enterprise (on their
Customer Edge [CE] routers), MPLS VPN-tobranch QoS is controlled by the service provider
(on their Provider Edge [PE] routers).
Branch CE
Central CE
MPLS
VPN
Branch CE
Service-Provider PE Routers
Therefore, to guarantee end-to-end QoS, enterprises

must comanage QoS with their MPLS VPN service
providers; their policies must be both consistent and
complementary.
QoS Design
for MPLS
VPN Subscribers
MPLS VPN service providers offer classes of service to
enterprise subscribers.
Admission criteria for these classes is the DSCP
markings of enterprise traffic. Thus, enterprises may
have to re-mark application traffic to gain admission
into the required service-provider class.
Some best practices to consider when assigning
enterprise traffic to service-provider classes of service
include:
Dont put Voice and Interactive-Video into the Real-Time
class on slow-speed ( 768 kbps) CE-to-PE links.
Dont put Call-Signaling into the Real-Time class on
slow-speed CE-to-PE links.
Dont mix TCP applications with UDP applications
within a single service-provider class (whenever
possible); UDP applications may dominate the class
when congested.
Example enterprise subscriber DSCP Remarking

Diagram and CE Edge Bandwidth Allocation Diagram.
Enterprise
Applications
DSCP
Routing
CS6
Voice
EF
Interactive-Video
AF41 CS5
Streaming-Video
CS4 AF21
DSCP 25 AF31
AF31
Call-Signalling
AF31/CS3 CS5
CS3
Transactional Data
AF21 CS3
Service-Provider
Classes of Service
EF
Real-Time
35%
CS5
CS6
Critical
20%
AF21
Video
15%
CS2
Network-Management
CS2
Bulk Data
AF11
Scavenger
CS1 0
Best-Effort
Bulk Data 5%
AF11/CS1
Best-Effort
25%
Outer-Circle: Enterprise Applications (CE Edge)

Inner-Circle: Service-Provider Classes
of Service (PE Edge)
Voice 15%
Best-Effort
24%
Enterprises should be aware of which RFC 3270

MPLS DiffServ Tunneling Mode their service
provider is using (Uniform Mode, Pipe Mode,
Scavenger
1%
or Short Pipe Mode). If Uniform Mode is
being used, the service provider may
Bulk Data 5%
re-mark customer traffic at Layer 3
NetworkManagement
(overriding the customer's DSCP markings).
2%
If such is the case, the enterprise subscriber
may need to restore such DSCP markings
Streamingon ingress (from the MPLS VPN).
Video 13%
Best-Effort
25%
Video
15%
Real-Time
35%
Critical
20%
Call-Signaling 5%
Routing 3%
Transactional
Data 5%
Mission-Critical
Data 12%
QoS Design for MPLS

VPN Service Providers
In order to support enterprise-subscriber voice,

video and data networks, service providers must
include QoS provisioning within their MPLS VPN
service offerings.
Service providers can mark at Layer 2

(MPLS EXP) or at Layer 3 (DSCP).
Shaded area represents service provider DiffServ domain.


This is due to the any-to-any/full-mesh nature of

MPLS VPNs, where enterprise subscribers depend
on their service providers to provision Provider-Edge
(PE) to Customer-Edge (CE) QoS policies consistent
with their CE-to-PE policies.

are based on
customer markings.
MPLS VPN
PE Router
CE Router
PE Router
CE Router
P Routers
In addition to these PE-to-CE policies, service

providers will likely implement ingress policers on
their PEs to identify whether traffic flows are in or
out-of-contract. Optionally, service providers may
also provision QoS policies within their core
networks, using Differentiated Services and/or
MPLS Traffic Engineering (TE).
IPP3/DSCP AF31
Packet initially
marked to IPP3/
DSCO AF31.
To guarantee end-to-end QoS, enterprises must

comanage QoS with their MPLS VPN service
providers; their policies must be both consistent
and complementary.
MPLS EXP 4
MPLS EXP 0
MPLS EXP 4
MPLS EXP 0
DSCP AF31
MPLS EXP 4
MPLS EXP values

from DSCP values.
DSCP AF31
Topmost label is
marked down by
a policer.
DSCP AF31
Topmost label is
popped and EXP
valueis copied to
underlying label.
DSCP AF31
Original customer-marked
IPP/DSCP values are
preserved.
Optional: Core DiffServ or MPLS TE Policies

PE Ingress Policing
and Remarking
P Routers
CE Router
PE Router
PE Router
CE Router
Required
Optional
MPLS VPN
PE-to-CE
LLQ/CBWFQ/WRED/
Shaping/LFI
RFC 3270 presents three modes of MPLS/DiffServ marking for service providers:
Uniform Mode: Service provider can re-mark customer DSCP values
Pipe Mode: Service provider does not re-mark customer DSCP values (service
provider uses independent MPLS EXP markings); final PE-to-CE policies are
based on service providers markings
Short Pipe Mode (shown above): Service provider does not re-mark customer
DSCP values (service provider uses independent MPLS EXP markings); final
PE-to-CE policies are based on customers markings
Service providers can guarantee service levels within their core by:
Aggregate Bandwidth Overprovisioning: Adding redundant links when utilization
hits 50%(simple to implement, but expensive and inefficient)
Core DiffServ policies: Simplified DiffServ policies for core links
MPLS TE: TE provides granular policy-based control over traffic flows within the core
cut here
cut here
QoS Design for IPSec VPNs

IPSec VPNs achieve network segregation and privacy
via encryption. IPSec VPNs are built by overlaying a
point-to-point mesh over the Internet using Layer 3encrypted tunnels. Encryption/ decryption is performed
at these tunnel endpoints and the protected traffic is
carried across the shared network.
2 Encryption/Decryption Delays
A marginal time element for encryption and decryption should be factored into the end-to-end delay budget
for Real-Time applications, such as VoIP. Typically these processes require 2-10 ms per hop but may be
doubled in the case of spoke-to-spoke VoIP calls that are homed through a central VPN headend hub.
e)
e)
Three main QoS considerations specific to

IPSec VPNs are
The additional bandwidth required by IPSec
encryption and authentication
The marginal time element required at each point
where encryption/decryption takes place
Anti-Replay interactions
Service
Provider
Si
Si
Campus
1 IPSec Bandwidth Overhead
CODEC
The additional bandwidth required to encrypt and

authenticate a packet needs to be factored into
account when provisioning QoS policies.
10-50 ms
(Depends on
Sample Siz
Branch
O
Offic
Office
Queuing
Variable
(Can Be Reduced
Using LLQs)
This is especially important for VoIP, where IPSec

could more than double the size of a G.729 voice
packet, as shown below.
Encrypt
Serialization
Propagation and
Network
Variable
(Can Be Reduced
Using LFI)
Minimal
2-10 ms
End-to-End Delay
(6.3 s / Km) +
Network Delay
(Variable)
Decrypt
Jitter Buffer
Minimal
2-10 ms
20-100 ms
(Depends on
Sample Siz
150 ms
3 Anti-Replay Interactions
Anti-Relay is a standards-defined mechanism to protect IPSec VPNs from hackers. If packets arrive outside of
a 64-byte window, they are considered compromised and are dropped prior to decryption. QoS queuing policies
may reorder packets such that they fall outside of the Anti-Replay window. Therefore, IPSec VPN QoS policies
need to be properly tuned to minimize Anti-Replay drops.
IPSec ESP
Tunnel Mode
G.729 VoIP
136 Bytes
G.729 VoIP
60 Bytes

Hdr
Hdr
Hdr IV
20
20
Packet 3 arrives out of order and

is outside the Anti-Replay window
IP
Hdr
UDP
RTP
64-Packet
Anti-Replay
Window
64 Packet
SlidingSliding
Window
Voice
69
GRE
IP
Hdr
UDP
RTP
Voice
20
12
20
ESP
ESP
Pad/NH Auth
2257
12
68
64
65
66
67
3
Anti-Replay
Drop
Packet 3 is considered compromised and

is not decrypted but is automatically
dropped by Anti-Replay
INDEX
Numerics
10-Class WAN Edge Model, 515
1P2Q1T queuing and dropping, 401404
1P3Q1T mode, 322
1PxQyT queuing, 36
2Q2T queuing and dropping, 396-398
show qos info config 2q2 tx verification
command, 398
show qos info runtime verification command,
399400
show queuing interface verification command,
400
33 percent limit (sum of LLQs), 450
4Q1T mode, 322
802.11e, 275277
802.1D, classes of service, 279
802.1Q/p, translating to and from DSCP, 9293
A
access control entries (ACEs), 225
access layer
Catalyst 3550 QoS design, 325
policers, 54
access point (AP), 271
access switches
configuring to conditionally trust CoS, 319
access switches (campus networks), 291
access-edge QoS design, 290
access-edge trust boundaries, 302
Conditionally Trusted Endpoint Models, 303,
307312
Trusted Endpoint Models, 302304
Untrusted Endpoint Models, 304307
access-edge utilization, 293
ACEs (access control entries), 225
ACLs, MQC-based class maps, 233
adaptive jitter buffers, 36
admission control, 197
admission criterion (Real-Time class), 563565

ADSL (Asynchronous Digital Subscriber Lines),
672
aggregate policers (Catalyst 6500), 258
aggregation routers, 505507
algorithms
MDRR, 585
queuing, 133
CBWFQ, 140
comparison, 138
PQ-WFQ, 139
priority queuing, 137
WFQ, 137
shaping, 120
token bucket algorithms, 105
analog gateways, 303
Anti-Replay drops, 638, 656
Anti-Replay functionality (IPSec QoS design),
654-656
any-to-any videoconferencing, 548549
AP (access point), 271
applications
data applications by class, 4647
Mission-Critical Data, 43
Streaming-Video, 41
unidirectional, 513
architectures (MPLS VPNs). See MPLS VPN-QoS
design
Assured Forwarding, 16
asymmetric links, 677
Asynchronous Digital Subscriber Lines (ADSL),
672
ATM, 176
PVC bundles, 147
Tx-rings, 489
WAN edge link-specific QoS design
ATM-FR SIW, 497501
high-speed links, 494495
medium-speed links, 493
slow-speed links, 488493
very-high-speed links, 496497
ATM CLP (ATM Cell-Loss Priority bit), 84
ATM inverse multiplexing over ATM, 493
ATM networks, 121122
ATM PVC bundles, slow-speed, 490492
714
ATM-FR SIW (ATM-to-Frame Relay Service Interworking)
ATM-FR SIW (ATM-to-Frame Relay Service

Interworking), 497498
attacks (worms), 50-54
authentication
ESP, 654
IPSec, 657
AutoQoS (Automatic QoS), 2425
Enterprise feature, 2526
evolution of, 28
B
backbone, 583587
bandwidth
Catalyst 4500, 251
Eight-Class Model, 461
guarantees, 137, 565
ISDN, 501
provisioning, 143, 645
Best-Effort class, 449
Real-Time class, 449
teleworker V3PN QoS, 674677
WAN aggregators, 449
VoIP, 646
reservations, 195
RSVP, 196197
statements, on class default, 457
VoIP streams, 3638
Bc (committed burst), 105, 479
Be (excess burst), 110, 480
bearer trafc
jitter, 3638
latency, 3435
loss, 34
Best-Effort class
bandwidth provisioning, 449
enabling WRED, 457
Best-Effort data, 44
best-effort networks, 11
best-effort service, 15
binary eponential backoff, 274
branch router QoS design, 513
case study, 535540
LAN edge, 517
branch-to-campus classication and

marking, 519525
DSCP-to-CoS remapping, 518
NBAR known worm classication and
policing, 526535
WAN edge, 514515
branch-to-branch trafc, 548
branch-to-campus classication and marking,
519521
NBAR application classification, 523525
source or destination IP address marking, 520
TCP/UDP classification, 522
broadband
serialization mitigation through TCP maximum
segment size tuning, 678679
split tunneling, 679681
UDP-based video applications, 678
broadband-access technologies, 671
cable, 673
DSL, 672
buffer space, 135
buffers, 3638
Bulk Data class, 44
C
cable, 671-673
DOCSIS 1.1 specification, 678
Integrated Unit + Access Models, 684685
overhead, 676677
uplink connections, 677
CAC (call admision control), 205
CallManager locations CAC, 209211
defined, 206
GK CAC, 211
local CAC tools, 208
measurement-based CAC tools, 208
prering CAC, 212
resource-based CAC tools, 209
RSVP, 212
tool categories, 207
VoIP CAC through RSVP, 215
Call-Signaling
MPLS VPN CE QoS design considerations, 553
campus QoS design, 295
Call-Signaling trafc, jitter, 38
case studies
CallManager environments, 296301

CallManager locations CAC, 209211
CallManager services, 295
campus Catalyst switches, 24
campus networks
oversubscription ratios, 291
QoS design. See campus QoS design
traffic, 339
underutilization, 290
access switches, 291
Call-Signaling, 295
case study, 422439
Catalyst 2950
Conditionally Trusted IP Phone + PC
Advanced Model, 322
Basic Model, 319320
queuing, 322325
Trusted Endpoint Model, 314315
Untrusted Multiapplication Server Model,
315319
Untrusted PC with SoftPhone Model, 315
Catalyst 2970/3750, 342343
Basic Model, 346348
enabling/disabling QoS, 343
queuing/dropping, 351356
Unconditionally Trusted IP Phone + PC
Basic Model, 348351
Untrusted Server Model, 345
Catalyst 3550, 325327
Advanced Model, 333336
Basic Model, 331333
queuing and dropping, 336341
Untrusted PC with SoftPhone Model,
327329
Catalyst 4500, 357358
715

Basic Model, 362364
queuing, 366370
show qos dbl command, 370
show qos interface command, 371372
show qos maps dscp tx-queue command,
370371
359360
Catalyst 6500, 372, 374
2Q2T queuing and dropping, 396400
CatOS defaults/recommendations, 375
Basic Model, 386387
congestion avoidance, 384
PFC3 distribution-layer Per-User
Microow Policing, 419
show port qos command, 376377
378382
defining designs, 292
DoS/worm mitigation, 292294
WAN aggregator/branch router handoff,
420422
campus-to-branch trafc, 548
CAR (committed access rate), 79, 107108
case studies
branch router QoS design, 535540
IPSec VPN QoS design, 686
telecommuter router, 694695
V3PN branch router design, 691693
VPN headend design, 687689
WAN aggregator QoS design, 689690
716
case studies
MPLS VPN QoS design, 616

CE routers, 617-619
PE routers, 620630
P routers, 630631
WAN aggregation router QoS design, 505507
Catalyst 2950, 314315
classification, marking, and mapping, 233
Conditionally Trusted IP Phone + PC Advanced
Model, 322
Conditionally Trusted IP Phone + PC Basic
Model, 319
CoS-to-DSCP map, 233
DSCP-to-CoS map, 234
policing and markdown, 234
queuing, 235, 322323
show wrr-queue bandwidth command, 324
show wrr-queue cos-map command, 325
range keyword, 315
315318
show class-map and show policy-map
verication commands, 318
show mls masks qos verication
command, 319
verication command, 318
vs. Catalyst 3550, 231
Catalyst 2970, 242, 342343
Model, 346348
queuing/dropping, 244246, 351356
Unconditionally Trusted IP Phone + PC Basic
Model, 348351
Catalyst 3550, 235, 325327
Model, 333336
Model, 331333
DSCP mutation maps, 237

queuing and dropping, 240241, 336339
show mls qos interface buffers verication
command, 340
show mls qos interface queuing
Catalyst 3750, 242, 342343
Model, 348351
Model, 347348
queuing/dropping, 244246, 351356
Catalyst 4500, 247, 357358
Model, 364366
Model, 362364
DSCP-to-queue maps, 251
enabling QoS, 248
queuing, 366370
370371
Catalyst 6500, 252, 372, 374
CatOS
commands

Basic Model, 386387
congestion avoidance, 384
defaults/recommendations, 375
378379, 381382
Model, 379382
classification, marking and mapping, 254256
enabling QoS, 254
PFC QoS, 261
PFC3 distribution-layer Per-User Microflow
Policing, 419
Supervisor 720, 253
Trusted Endpoint Model
VLAN-based QoS, 254
WRED-drop thresholds, 262
Catalyst 6500 conguring microow policers, 257
Catalyst QoS Models, 224
classification, 225
policing, 227
queuing, 228-230
CatOS defaults/recommendations, 375
CBR (constraint-based routing), 603
CBWFQ (Class-Based Weighted Fair Queuing),
133, 139140
CDP (Cisco Discover Protocol), 307
CE bit, 163
CE design, 556563
CE routers, MPLS VPN QoS design case study,
617-619
CEF (Cisco Express Forwarding), 74
Channel Utilization eld, 278
CIR (committed information rate), 105
Frame Relay networks, 479
policing behavior based on percentages, 118
Cisco 12000 routers
priority command, 587
queuing, 585
717
Cisco Discover Protocol (CDP), 307

Cisco Express Forwarding (CEF), 74
class default policing, 112
class selectors, 16
class-based Frame Relay trafc shaping, 123124
class-based marking, 77
class-based policing, 79, 109
benefits, 109
single-rate three-color marker/policer, 110112
two-rate three-color marker/policer, 112113
class-based shaping, 126-127
Class-Based Weighted Fair Queuing (CBWFQ), 133
classes of service (802.1D), 279
classication, 57, 6869
branch-to-campus, 519
Catalyst 2950, 233
Catalyst 2970, 243
Catalyst 3550, 237
Catalyst 3750, 243
Catalyst 4500, 248249
Catalyst 6500, 254256
Catalyst QoS models, 225
NBAR application, 523524
source or destination IP addresses, 520
TCP/UDP, 522
tools, 70
MQC-based class maps, 7172
NBAR, 7377
Code Red, 527
codecs (frame-based), 34
CodeRedv2, 528
color-aware policing, 116117
color-blind policing, 117
commands, 4748
commit all command, 377
dbl policy command, 367
frame-relay fragment command, 481
match protocol dlsw command, 48
max-reserved-bandwidth, 559563
mls qos cos override interface command, 316
ping vrf command, 601602
ppp multilink links minimum command, 504
priority-queue out command, 240
qos dbl command, 367
qos map dscp to tx-queue command, 367
show atm bundle command, 493
show atm pvc command, 489
show atm vc command, 492
718
commands
show class-map verification command, 318

show controllers command, 451
show frame-relay fragment command, 482
show ima interface atm command, 495
show ip access-list command, 521
show ip bgp vpnv4 all command, 615
show ip nbar map command, 525
show ip rsvp interface command, 599
show ip rsvp neighbor command, 599
show mls masks qos verification command, 319
show mls qos command, 358
show mls qos interface buffers verification
command, 340
show mls qos interface policers verification
command, 318
show mls qos interface queuing verification
command, 341
show mls qos interface statistics verification
command, 329
show mls qos maps command, 356
show mls qos maps dscp-output-q command,
356
show mpls interface command, 600
show mpls traffic-eng topology command,
614615
show mpls traffic-eng tunnels command, 601
show mpls traffic-eng tunnels summary
command, 600
show policy command, 456
show policy interface command, 469
show policy interface verification command,
329
show policy-map verification command, 318
show port qos commands, 376377
show ppp multilink command, 478
show qos acl verification command, 380
show qos command, 358
show qos info config 2q2 tx verification
command, 398
show qos info runtime verification command,
399400
370371
show qos maps verification command, 379
show qos policer verification command, 381
show qos statistics verification command, 382

show queuing interface verification command,
400
trust-device command, 386
tx-queue command, 367
tx-ring-limit command, 490
wrr-queue bandwidth command, 322
wrr-queue cos-map command, 240
wrr-queue dscp-map interface configuration
command, 339
wrr-queue queue-limit command, 240
wrr-queue queue-limit interface command, 338
show mls qos map, 320
commit all command, 377
committed access rate (CAR), 79, 107108
committed burst rate (Bc, 105, 479
committed information rate. See CIR
compatibility (802.1D classes of service), 279
compression
G.729 voice compression, 170
hardware compression, 181
Conditionally Trusted Endpoint Models (Trust
Boundaries), 303, 307312
Model, 322
Catalyst 2970/3750, 348351
Catalyst 3550, 333336
Catalyst 4500, 364366
Catalyst 6500, 387391
Conditionally Trusted IP Phone + PC Basic Model,
319, 346347
Catalyst 2970/3750, 346348
Catalyst 3550, 331333
Catalyst 4500, 362364
Catalyst 6500, 386387
conguring
1P2Q2T queuing, 407
1P3Q1T queuing, 410
1P3Q1T queuing model, 337
1P3Q8T queuing, 412
1P7Q8T queuing, 416417
Catalyst 2950 switches
Advanced Model, 322
custom queuing (CQ)

Basic Model, 319320
queuing, 322323
315316, 318319
CoS-to-queue mapping (Catalyst 3550), 240
cRTP for ATM links, 176
DSCP mutation, 238
FR-VATS, 125
individual policer on Catalyst 4500, 250
IPSec authentication, 657
MCMP for an ISDN interface, 185
microflow policing on Catalyst 6500, 257
MLP LFI, 184
MPLS DS-TE, 606613
MPLS per-VPN TE, 592598
PFC, 255
policing, 234
QoS on Cisco APs, 281
queuing (Catalyst 2950), 235
RSVP, 196
SRR shaping and sharing weights on Catalyst
2970/3750, 246
trust on Catalyst 6500, 255
VLAN-based QoS on Catalyst 6500, 254
WRED, 162
conguring WRED-drop thresholds, 262
conrming trafc, 107
congestion avoidance
Catalyst 6500, 384
tools, 159
DSCP-based WRED, 162163
explicit congestion notication, 163
RED, 160
WRED, 161162
congestion-management tools, 133
converged networks, 135136
connecting trusted endpoints, 304
consistent QoS behavior, 453
constraint-based routing (CBR), 603
constricted channels, 159
control plane QoS
IP routing, 4849
network management, 49
719
controlled load, 15, 197

controlling trafc
branch-to-branch, 548
campus-to-branch, 548
converged networks
congestion-management tools, 135136
QoS, 1214
convergence, 449
Core Best-Effort class, 584
Core Critical Data class, 584
core QoS considerations, 582
aggregate bandwidth overprovisioning, 583
platform specic considerations, 585587
Three-Class Provider-Core Model,
583585
MPLS traffic engineering, 587
basic, 588590
MPLS DS-TE, 603615
Core Real-Time class, 584
CoS values, assigning queues, 240
CoS-to-DSCP maps
Catalyst 2950, 233
Catalyst 6500, 256
control plane provisioning, 657
CQ (custom queuing), 137
cRTP (RTP header compression), 172
class-based header compression, 178179
configuring for ATM links, 176
formats, 173
Cisco propriety format, 173
IETF format, 174
IPHC, 173
formats and encapsulation summary, 177178
incompatibility with IPSec, 643644
Layer 2 encapsulation protocol support, 175
ATM, 176
Frame Relay, 176
HDLC, 175
PPP, 175
LLQ, 145
policing and shaping, 180
tunnels, 180
crypto engine, 651652
cTCP (TCP header compression), 171
custom queuing (CQ), 137
720
CWmax
CWmax, 278, 282

CWmin, 274, 278, 282
D
data
applications by class, 4647
QoS, 4243
DLSw+, 4748
locally dened Mission-Critical Data, 45
Transactional Data/Interactive Data, 45
data frames (802.11), 272
data VLANs (DVLANs), 314
datagrams, 153
data-link connection identiers (DLCIs), 123
data-link switching plus (DLSw+), 4748
DBL (dynamic buffer limiting), 366-367
DCF (Distributed Coordination Function), 272
Interframe Spaces, 272
random backoffs, 273
DDR (dial-on-demand routing), 503
DDT (delay to dial tone), 38
delay budgets (IPSec VPNs), 647
delay to dial tone (DDT), 38
delay variation, 1314. See also jitter
deploying
IPSec VPNs via DMVPN, 646
LFI tools, 450
policers, 106
QoS designs, 62
Untrusted Server Model on Catalyst 2950, 315
designing QoS
classification and marking principles, 57
deployment, 62
DoS and worm mitigation principles, 6162
policing and markdown principles, 5758
queuing and dropping principles, 5860
destination IP address classication, 520
DHCP, translating to Frame Relay DE bit, 94
dial-on-demand routing (DDR), 503
Differentiated Services code points (DSCPs), 87
DiffServ, 16
advantages of DiffServ model, 16
deploying in backbone
platform specic considerations, 585587
Three-Class Provider-Core Model,

583585
DIFS, 272
Digital Subscriber Line. See DSL
disabling
flow control, 327
native DLSw+ ToS markings, 48
QoS on Catalyst 2970/3750, 343
Discard class placeholder, 568
Distributed Coordination Function (DCF), 272
distributed platform frame relay links, 486-487
distributed platform QoS, 453
distributed trafc shaping (DTS), 128, 486
Distributed-Platform/Consistent QoS Behavior QoS
Baseline Model, 465466
distribution layer, Catalyst 3550 QoS design, 325
DLCIs (data-link connection identiers), 123
dlsw tos disable command, 48
dlsw tos map command, 47
DLSw+ (data-link switching plus), 4748
DMVPNS (Dynamic Multipoint Virtual Private
Networks), 646
DOCSIS 1.1 specication, 674, 678
dominating links (VoIP), 449
DoS attacks
campus network mitigation strategies, 292294
mitigation principles, 6162
downstream QoS, 271
drop thresholds (Catalyst 2970 and 3750), 245
dropping, 5860
Anti-Replay, 656
Catalyst 2970, 244246
Catalyst 3550, 240241, 336339
Catalyst 3750, 244246
Catalyst 4500, 250252
Catalyst 6500, 259263, 391394
DSCPs (Differentiated Services code points), 87
mutation maps (Catalyst 3550), 237
DSCP-to-CoS maps
Catalyst 2950, 234
Catalyst 3550, 237
Frame Relay trafc shaping (FRTS)

DSCP-to-queue maps
Catalyst 4500, 251
DSL (Digital Subscriber Line), 671-672
Integrated Unit + Access Models, 684685
Integrated Unit/Dual-Unit models, 682
uplink connections, 677
DSL (AAL5 + PPPoE) overhead, 675676
DSLAM (DSL Access Multiplexer), 671
DTS (distributed trafc shaping), 128, 486
Dual-Unit Model, 669, 682
DVLANs (data VLANs), 314
dynamic buffer limiting (DBL), 366
Dynamic Multipoint Virtual Private Networks
(DMVPNs), 646
E
EAP (Extensible Authentication Protocols), 308
ECN bit, 164165
ecn keyword, 165
ECT bit, 163
EDCF (Enhanced Distributed Coordination
Function), 275277
EI (Enhanced Image), 232
Eight-Class Model, 460-462
Eight-Class Site-to-Site V3PN Model, 660664
EMI (Enhanced Multilayer Software Image), 243
enabling
MLPoATM, 499
QoS
Catalyst 4500, 248
Catalyst 6500, 254
Catalyst 2970/3750, 343
encryption, delay budgets, 648
end users network expectations, 9
endpoints, 201, 304
end-to-end QoS, 10
Enhanced Distributed Coordination Function
(EDCF), 275, 277
Enhanced Image (EI), 232
Enhanced Multilayer Software Image (EMI), 243
enterprise resource planning (ERP), 42
ERP (enterprise resource planning), 42
errors (Anti-Replay), 655
ESP authentication, 654
721
Ethernet 802.1Q tunnels, 82

Ethernet 802.1Q/p, 8182
Ethernet downstream, 271
evolution of QoS, 29
exceeding trafc, 107
excess burst rate (Be), 110, 480
expedited forwarding, 16
explicit congestion notication, 163
Extensible Authentication Protocols (EAP), 308
F
FIFO Tx-ring, 152
Five-Class Model, 456459
Five-Class Provider-Edge Model, 565566
MPLS VPN CE QoS design considerations,
561563
Fixed Slot Time Default values, 278
ow control, disabling, 327
Four-Class Provider-Edge Model, 565
559561
fragment sizes
distributed platform Frame Relay links, 486
WAN link fragmentation, 183184
frame-based codecs, 34
Frame Relay
cRTP, 176
Frame-Relay fragmentation, 185
FRF.11.1 and FRF.12.1, 187188
LFI for Frame Relay/ATM service
interworking, 188189
PVCs, 186187
PIPQ, 150
WAN edge link-specific QoS design, 478
Bc, 479
Be, 480
CIR, 479
distributed platform links, 486487
Frame Relay bundles, 148
Frame Relay DE bit, translating to from DSCHP, 94
Frame Relay Dual-FIFO, 150
Frame Relay trafc shaping (FRTS), 122123
722
Frame Relay voice-adaptive trafc shaping (FR-VATS)
High Link-Speed QoS Class Models, 459

Distributed-Platform/Consistent QoS Behavior
QoS Baseline Model, 465466
Eight-Class model, 460-462
High-Level Data Link Control (HDLC), 135
high-speed ATM links, 494495
high-speed frame relay links, 484485
high-speed leased lines, 472476
pkts matched statistics, 477
horizontal separation of trafc, 107
how qos dbl command, 370
hub routers
hub-and-spoke topology, 548, 646
Frame Relay voice-adaptive trafc shaping

(FR-VATS), 124125
Frame-Relay DE bit, 83
frame-relay fragment command, 481
FRF.11.1 and FRF.12.1
fragmenting, 187188
FRF.8, 189
FRTS (Frame Relay trafc shaping), 122123
FR-VATS (Frame Relay voice-adaptive trafc
shaping), 124125
G-H
G.729 voice compression, 170
G.SHDSL, 673
gatekeepers (GK), 211
generic trafc shaping, 126
GK (gatekeepers), 211
GK CAC, 211
global synchronization, 159
goals of convergence, 449
guaranteed load service, 197
guaranteed services, 15
guarantees (bandwidth), 137, 195, 565
handoffs (WAN aggregator/branch router), 420422
hardware compression, 181
hardware crypto engines, 652
HDLC (High-Level Data Link Control), 135, 175
header-compression techniques, 170
formats,
IETF format, 174
IPHC, 173
ATM, 176
Frame Relay, 176
HDLC, 175
PPP, 175
RTP header compression (cRTP), 172
standards, 171
TCP header compression (cTCP), 171
hierarchical class-based shaping, 127
hierarchical policing, 114
I
IANA (Internet Assigned Numbers Authority), 522
IETF (Internet Engineering Task Force), 7
IETF format, 174
IMA (ATM inverse multiplexing over ATM), 493
Integrated Services, 6
Integrated Unit + Access Model, 669670, 684685
Integrated Unit Model, 668, 682
Interactive Data, 45
Interactive-Video, 39
Interframe Spaces, 272
Internal DSCP value, 225
Internet Assigned Numbers Authority (IANA), 522
interoperability (RSVP), 213
IntServ, 7, 15
IP conguring stations, 303
IP header compression format (IPHC), 171173
IP Precedence, 567
IP routing, 4849
ip rsvp bandwidth command, 215
IP RTP header compression, 451
IP RTP priority, 139
IP telephony, 307
IP ToS (IP type of service), 8687
IP VPN Multiservice, 551
IPHC (IP header compression format), 171173
Layer 2
IPSec
authentication, 657
incompatibility with cRTP, 643644
LLQ, 145
prefragmentation, 190
IPSec-encrypted G.729 packets, 642
IPSec Encryption Engines, 652
IPSec QoS design, 635
Anti-Replay functionality, 655
anti-replay functionality, 654656
control plane provisioning, 657
cRTP and IPSec incompatibility, 643644
delay budget increases, 647
headend VPN edge QoS options for site-to-site
V3PNs, 665666
packet overhead increases, 640642
pre-encryption queuing, 651653
QoS Pre-Classify, 649
site-to-site V3PN, 637
IPSec transport mode (encrypting an IP
GRE tunnel), 638
IPSec tunnel mode (encrypting an IP GRE
tunnel), 639640
IPSec tunnel mode (No IP GRE tunnel),
638
site-to-site V3PN QoS models
Eight-Class Site-to-Site V3PN Model,
660-664
Six-Class Site-to-Site V3PN Model,
658659
asymmetric links and unidirectional QoS,
677
deployment models, 667670
topologies, 646
ToS byte preservation, 649
VPNs, 635
IPSec transport mode (encrypting an IP GRE
tunnel), 638
IPSec tunnel mode (encrypting an IP GRE tunnel),
639640
IPSec tunnel mode (No IP GRE tunnel), 638
723
IPSec VPN QoS design (case study), 686

telecommuter router, 694695
V3PN branch router design, 691693
VPN headend design, 687689
WAN aggregator QoS design, 689690
ISDN, 671
WAN edge link-specific QoS design
CallManager CAC limitations, 503
MLP packet reordering, 502
variable bandwidth, 501
voice and data on multiple ISDN B
channels, 503504
ITDP/UDP ports (CallManager environments),
296301
J-K
jitter, 13, 35, 450
jitter buffers, 3738
adaptive, 36
underruns, 14
keywords, 358
L
LAN edge QoS design, 517
branch-to-campus classification and marking,
519525
NBAR known worm classification and
policing, 526535
LANs
switching environments, 223
QoS for wired vs. wireless, 270
latency
VoIP, 3435
Layer 2
access (MPLS VPN CE QoS design), 550551
marking elds, 8182
ATM CLP, 84
Frame-Relay DE bit, 83
MPLS EXP bits, 84
queuing mechanisms, 150
queuing subsystems, 136
724
Layer 3 marking elds
Layer 3 marking elds, 88-90

Layer 3 queuing mechanisms
CBWFQ, 139140
legacy, 136137
LLQ, 133, 140, 199, 450, 652- 653
ATM PVC bundles, 147
cRTP, 145
IPSec, 145
LFI, 147
MLP and Frame Relay bundles, 148
operation, 141
policing, 142
VoFR, 149
Layer 3 queuing subsystems, 135
leased lines, 467
high-speed, 472478
medium-speed, 471
slow-speed, 467470
legacy Layer 3 queuing mechanisms, 136137
LFI (Link Fragmentation and Interleaving)
for Frame Relay/ATM service interworking,
188189
LLQ, 147
tools, 182, 450
line card queuing structures (catalyst 6500),
393396
links
asymmetric, 677
ATM
high-speed, 494495
medium-speed, 493
slow-speed, 488493
very-high-speed, 496497
capacity, 293
Eight-Class Site-to-Site V3PN Model, 662
Frame Relay networks
distributed platform, 486487
high-speed, 484485
medium-speed, 482-484
slow-speed, 480482
speed, 452
VoIP, dominating, 449
link-specic tools, 19, 169
LLQ (low-latency queuing), 133, 140, 199, 450,
652-653
ATM PVC bundles
LLQ, 147

cRTP, 145
IPSec, 145
LFI, 147
MLP and Frame Relay bundles, 148
operation, 141
policing, 142
VoFR, 149
VoIP and multiple levels of data, 141
local CAC tools, 208
loss (voice), 34
low link speeds (WANs), 450
low-latency queuing. See LLQ
LS VPN QoS design, 613615
M
mapping
Catalyst 2950, 233
Catalyst 2970, 243
Catalyst 3550, 237
Catalyst 3750, 243
Catalyst 4500, 248249
Catalyst 6500, 254256
IP Precedence, 567
Mapping Models (enterprise-to-service
provider)
Five-Class Provider-Edge Model,
565566
Three-Class Provider-Edge Model,
563564
markdown, 5758
Catalyst 2950, 234
Catalyst 2970, 244
Catalyst 3550, 238239
Catalyst 3750, 244
Catalyst 4500, 249250
Catalyst 6500, 257259
markers (policers as), 107
marking, 57, 6869
branch-to-campus, 519
Catalyst 2970, 243
Catalyst 3550, 237
MPLS VPN QoS design
Catalyst 3750, 243

Catalyst 4500, 248249
Catalyst 6500, 254256
DLSw+ traffic, 48
554556
tools
class-based marking, 77
class-based policing, 79
Layer 2 marking elds, 8184
Layer 3 marking elds, 8687
Layer 3 tunnel marking tools, 8890
translating Layer 2 and Layer 3 packet
markings, 9098
voice gateway packet marking, 7981
traffic, 304
match protocol commands, 75
match protocol dlsw command, 48
max-reserved-bandwidth command, 559563
max-reserved-bandwidth interface command, 143
MCMP (Multiclass Multilink PPP), 185
MDRR (modied-decit round-robin) algorithm,
585
mean opinion scores (MOS), 35
measurement-based CAC tools, 208
Media Gateway Control Protocol (MGCP), 79
Medium Link-Speed QoS Class Models, 454
medium-speed ATM links, 493
medium-speed frame relay links, 482484
medium-speed leased lines, 471
MGCP (Media Gateway Control Protocol), 79
Microow policers, 257
Mission-Critical Data applications, 43-45
mitigating serialization delay, 678
MLP (Multi Point-to-Point Protocol), 136
MLP bundles, 148
MLP LFI (Multilink PPP Link Fragmentation and
Interleaving), 183185
MLP packets, reordering, 502
MLPoATM, 488489, 499
MLPoFR (MLP over Frame Relay), 177
mls prex keyword, 358
mls qos cos override command, 316
modied-decit round-robin (MDRR) algorithm,
585
modular QoS CLI based class maps, 7172, 233
725
Modular QoS Command-Line Interface (MQC),

1920
MOS (mean opinion scores), 35
MPLS DiffServ Tunneling modes, 566
Pipe Mode, 573582
Short Pipe Mode, 569573
Uniform Mode, 567569
MPLS EXP bits, 84
MPLS Trafc Engineering, 199, 587
basic, 588590
MPLS DS-TE, 603605
conguring, 606612
P-router conguration, 612613
show mpls trafc-eng topology command,
614
ping vrf tunnels command, 601602
show mpls trafc-eng tunnels command,
601
command, 600
MPLS VPN CE QoS design
special considerations, 550-552
561563
Four-Class Provider-Edge Model,
559561
Layer 2 access, 550551
marking/re-marking, 554556
service-provider service-level agreements,
551
TCP and UDP, 553554
556559
voice and call signaling, 553
voice and video, 553
MPLS VPN QoS design, 547
case studies, 616
CE routers, 617619
PE routers, 620630
P routers, 630631
726
MPLS VPN QoS design
core considerations, 582

aggregate bandwidth overprovisioning,
583
DiffServ in the backbone, 583587
MPLS trafc engineering, 587612
need for QoS, 548550
MQC (Modular QoS Command-Line Interface),
19-20
MQC/ACL classication, 233
MQC-based class maps, 7172
ACLs, 233
multiaction policing, 115
Multi Point-to-Point Protocol (MLP), 136
Multiclass Multilink PPP (MCMP), 185
multilink fragment-delay 10 command, 185
Multilink PPP Link Fragmentation and Interleaving
(MLP LFI), 183185
multiple priority classes, 72
N
NAT transparency feature overhead, 675
NBAR (Network-Based Application Recognition),
25, 72
application classication, 523524
known-worm classication and policing, 526
Code Red, 527
CodeRedv2, 528
future worms, 533534
NIMDA, 529
policing worms, 534535
RPC DCOM/W32MS Blaster, 531532
Sasser worm, 532533
SQL Slammer, 530
Packet Description Language Modules
(PDLMs), 520
protocol classification, 7476
RTP payload classification, 77
NBAR exchange PDLM, 532
NBAR netbios PDLM, 532
NBMA (nonbroadcast multiaccess), 119
nested hierarchical policing, 115
Network-Based Application Recognition. See
NBAR
networks
Best-Effort, 11
end user expectations, 9
management (QoS), 49
VoIP design considerations, 34
NIMDA, 529
nonbroadcast multiaccess (NBMA), 119
O-P
out-of-prole trafc, 227
overprovisioning LLQ trafc, 450
P routers, 549
packets, 18
MLP, reordering, 502
overhead increases (IPSec QoS design),
640642
packetization delay, 13
PAK_priority, 153, 452, 657
PAK_priority ag, 49
PBR (policy-based routing), 79
PBS (peak burst size), 112
PDLMs (NBAR Packet Description Language
Modules), 74, 520
PE QoS considerations, 563
Enterprise-to-Service Provider Mapping
Models
565566
Mapping Models, 563
563564
MPLS DiffServ Tunneling modes, 566
Pipe Mode, 573582
PE routers, 620630
peak burst size (PBS), 112
peak information rate (PIR), 112, 118
peak rate, 121
peak-rate shaping, 121
percent keyword, 140
percentage-based policing, 116
QoS
percentage-based shaping, 127

performance (cRTP), 181
per-Port/per-VLAN policing, 239
PE-to-P design, 583
PFC, conguring, 255
PFC3, 253
PFC3 distribution-layer Per-User Microow
Policing (Catalyst 6500), 419
PIFS, 272
ping vrf command, 601602
Pipe Mode, 573-582
PIPQ (PVC Interface Priority queuing), 150
PIR (peak information rate), 112, 118
pkts matched statistics, 477
placeholders. 568
PoC (proof-of-concept) tests, 62
police statements, 72
policers, 103, 107
CAR, 107108
class-based, 109
benets, 109
single-rate three-color marker/policer,
110112
two-rate three-color marker/policer,
112113
color-aware policing, 116117
color-blind policing, 117
compared to shapers, 104
default, 117
deploying, 106
DoS/worm mitigation (campus networks), 293
hierarchical policing, 114
as markers, 107
microflow, 257
multiaction policing, 115
percentage-based policing, 116
policies
access switches, 291
LAN switching environments, 223
on P routers, 549
on routers, 549
policing, 5758
Catalyst 2950, 234
Catalyst 2970, 244
Catalyst 3550, 238239
Catalyst 3750, 244
Catalyst 4500, 249250
Catalyst 6500, 257259
727

class-based, 79
cRTP, 180
LLQ, 142
worms, 534535
policy-based routing (PBR), 79
policy-map, 20
porting software QoS to hardware, 223
ports
presetting those used by SoftPhone, 315
trust states, 225
PPP, 175
ppp multilink links minimum command, 504
PPPoFR (PPP over Frame Relay), 177
PQ (priority queuing), 137
PQ-WFQ, 137-139
pre-encryption queuing, 651653
IPSec transport mode, 639
IPSec tunnel mode, 638
prering CAC, 212
prioritization, 679
priority classes, police statements, 72
priority-queue out command, 240
propagation delay, 13
protecting video, 557
Protocol Description Language Module (PDLM),
74, 520
P routers, 630631
provisioning (bandwidth), 645-646
proxies, 201
P-to-P design, 583
PVC Interface Priority queuing (PIPQ), 150
PVCs
bundling, 492
fragmenting, 186187
VoFR, 188
Q
QBSS IE (QoS basic service set information
element), 278
QoS
access-edge design, 290
branch routers, 513514
728
QoS
campus networks. See campus QoS design

Catalyst Models. See Catalyst QoS models
Cisco APs, 280281
classification and marking principles, 57
control plane
IP routing, 4849
network-management, 49
data, 4243
DLSw+, 4748
Transactional Data/Interactive Data, 45
deploying, 62
design principles, 55
DiffServ, 16
disabling on Catalyst 2970/3750, 343
DoS and worm mitigation principles, 6162
enabling on Catalyst 2970/3750, 343
end-to-end, 10
evolution of, 78, 26, 29
guidance, 27
historical perspective, 56
IntServ, 15
link-specific tools, 19, 169
models, 14
need for on MPLS VPNs, 548550
network expectations of end users, 9
policies required on WAN aggregators, 448
policing and markdown principles, 5758
porting software QoS to hardware, 223
queuing and dropping principles, 5860
Scavenger class, 49-54
simplifying, 19
AutoQoS, 2426
cross-platform feature consistency, 24
default behavior, 21
MQC, 20
QoS Baseline, 2022
tool set, 1718, 223
upstream vs. downstream, 271
video, 39
interactive, 39
streaming, 41
VoIP, 33
bearer trafc, 3438
Call-Signaling trafc, 38
WAN edge link-specific

ATM, 488497
ATM-FR SIW, 497501
Frame Relay, 478487
ISDN, 501504
leased lines, 467478
wireless LANs vs. wired LANs, 270
QoS Baseline Model, 20-21, 463465
class deployment, 55
QoS design principles, 55
recommendations, 22
QoS basic service set (QBSS), 278
qos dbl command, 367
QoS Design Guide, 27
QoS group placeholder, 568
qos map dscp to tx-queue command, 367
QoS Pre-Classify, 649
QoS preclassify feature, 89
queuing, 5860, 133134
algorithms
CBWFQ, 140
comparison, 138
PQ-WFQ, 139
WFQ, 137
buffer space, 135
Catalyst 2950, 235
Catalyst 2950 switches, 322323
Catalyst 2970, 244246, 351356
Catalyst 3550, 240241, 336339
command, 340
show mls qos interface queuing
Catalyst 3750, 244246
Catalyst 4500, 250252, 366-370
370371
Catalyst 6500, 259263, 391394
RTP payload classication

l2Q2T queuing and dropping, 396400
line card queuing structures, 393396
Catalyst QoS models, 228, 230
Cisco 12000 routers, 585
default queue limits, 338
Layer 2 queuing mechanisms, 150
Layer 3 queuing mechanisms
CBWFQ, 139140
legacy, 136137
LLQ, 140149
LLQ, 652
policies on routers, 549
reducing queue limits, 657
software (WAN aggregators), 448449
transmit ring (Tx-ring), 136
Tx-ring, 152
queuing tools, 133
R
radio downstream QoS, 271
radio upstream QoS, 271
RAI (resource activity indicator), 209
random backoffs, 273
random-detect command
ecn keyword, 165
Real-Time class
admission criterion, 563565
RED (Random Early Detection), 160
re-marking
554556
traffic, 304
reservations, 196197
resource activity indicator (RAI), 209
resource-based CAC tools, 209
RFC 2205, 195
RFC 2597, 58
RFC 3168, 163
RFC 3246, 36
ROHC (robust header compression), 171
routers
branch routers, 447
hub routers, 548
729
P routers, 549
policies, 549
roles in WANs, 447
link speeds, 452
PAK_priority, 452
required QoS policies, 448
serialization, 450
software queuing, 448449
Tx-ring tuning, 451
routing
DDR, 503
packets-per-second capability, 651
RPC DCOM/W32/MS Blaster, 531532
RSVP, 195
admission control, 197
CAC, 212
configuring, 196
cRTP, 180
interoperability, 213
LLQ, 199
overview, 196
scalability, 199
security, 213
service types, 197
VoIP CAC through RSVP, 215
RSVP-DiffServ integration, 200
RSVP PATH message, 196
RSVP RESV message, 196
RTP header compression (cRTP)
formats, 173
IETF format, 174
IPHC, 173
formats and encapsulation summary, 177178
Frame Relay, 176
HDLC, 175
PPP, 175
policing and shaping, 180
tunnels, 180
RTP payload classication, 77
730
SAR (Segmentation and Reassembly) engine
S
SAR (Segmentation and Reassembly) engine, 675
SAs (security associations), 638
Sasser worm, 532533
scalability
IPSec VPN QoS design case study, 686
RSVP, 199
Scavenger class
DoS and worm mitigation, 5054
QoS, 49
Scavenger-class QoS strategy, 294
SCCP (Skinny Call Control Protocol), 295
scheduling tools, 133-134
SCSP mutation maps (Catalyst 6500), 257
security
RSVP, 213
worms, 50
security associations (SAs), 638
Serial Line IP (SLIP) protocol, 173
serialization, 678
delay, 13
servers, 303
service provider service-level agreements, 551
service types (RSVP), 197
service-policy, 20
services for CallManagers, 295
shapers, 103, 118
ATM networks, 121122
class-based Frame Relay traffic shaping,
123124
class-based shaping, 126127
compared to policers, 104
cRTP, 180
distributed traffic shaping (DTS), 128
Frame Relay traffic shaping (FRTS), 122123
Frame Relay voice-adaptive traffic shaping,
124
generic traffic shaping, 126
peak-rate shaping, 121
shaping algorithms, 120
show class-map verication command, 318
show controllers command, 451

show frame-relay fragment command, 482
show ima interface atm command, 495
show ip access-list command, 521
show ip nbar port-map command, 525
show mls masks qos verication command, 319
show mls qos command, 358
command, 340
show mls qos interface policers verication
command, 318
show mls qos interface queuing verication
command, 341
show mls qos interface statistics verication
command, 329
show mls qos interface verication command, 314
show mls qos maps command, 356
show mls qos maps dscp-output-q command, 356
show mpls trafc-eng topology command, 614
show mpls trafc-eng tunnels command, 601
show mpls trafc-eng tunnels summary command,
600
show policy command, 456
show policy interface command
high-speed leased lines, 473
slow-speed leased lines, 469
show policy interface verication command, 329
show policy-map interface command, 178
show policy-map verication command, 318
show port qos commands, 376377
show qos acl verication command (Catalyst 6500),
380
show qos command, 358
show qos info cong 2q2 tx verication command,
398
show qos info runtime verication command,
399400
show qos maps dscp tx-queue command, 370371
show qos maps verication command (Catalyst
6500), 379
ToS (type of service)
show qos policer verication command (Catalyst

6500), 381
show qos statistics verication command (Catalyst
6500), 382
show queuing interface verication command, 400
SI (Standard Image), 232
SIFS, 272
site-to-site V3PN, 637
headend VPN edge QoS options, 665666
IPSec transport mode (encrypting an IP GRE
tunnel), 638
IPSec tunnel mode (encrypting an IP GRE
tunnel), 639640
IPSec tunnel mode (No IP GRE tunnel), 638
QoS models
Eight-Class Site-to-Site V3PN Model,
660-664
Six-Class Site-to-Site V3PN Model,
658659
Six-Class Site-to-Site V3PN Model, 658659
Skinny Call Control Protocol (SCCP), 295
SLIP (Serial Line IP) protocol, 173
Slow Link-Speed QoS Class Models, 454
slow-speed ATM links, 488489
ATM PVC bundles, 490492
Tx-rings, 490
slow-speed Frame Relay links, 480481
slow-speed leased lines, 467469
show interface command, 469
slow-speed links (ATM-FR SIW), 499501
SMI (Standard Multilayer Software Image), 243
SoftPhone, 315
software queuing (WAN aggregators), 448449
source IP address classication, 520
speed (links), 452
SQL Slammer, 530
Standard Image (SI), 232
Standard Multilayer Software Image (SMI), 243
state-machine synchronization, 212
streaming video, 41, 557
731
strict-priority queuing rule, 59

sum of LLQs, 450
Supervisor 720, 253
T
table map feature, 98
tail drops, 241
TCP
global synchronization behavior, 159
packet loss, 656
and UDP, 553554
TCP/UDP classication, 522
TAM (time-division multiplexing), 105
asymmetric links and unidirectional QoS, 677
cable overhead, 676677
DSL (AAL5 + PPPoE) overhead, 675676
NAT transparency feature overhead, 675
broadband serialization mitigation through TCP
maximum segment size tuning, 678679
cable, 673
DSL, 672
business-ready teleworker design, 666
Deployment Models, 667, 682
Dual-Unit Model, 669
Integrated Unit + Access Model, 669670.
684-685
Integrated Unit Model, 668
Integrated Unit/Dual Unit Models,
682-684
Three-Class (Voice and Data) Model, 454-456
Three-Class Provider-Core Model, 583
Three-Class Provider-Edge Model, 556559,
563564
time-division multiplexing (TDM), 105
token bucker algorithms, 105
topologies
split tunnel, 680
ToS (type of service), 47
byte preservation, 649
reflection, 90
732
total drops statistics
total drops statistics, 477

trafc
branch-to-branch, 548
campus networks, 339
campus-to-branch, 548
classification, 6877
conforming, 107
data, 42
defined by QoS Baseline, 21
DLSw+, marking, 48
exceeding, 107
handoffs, 421
horizontal separation of, 107
IP, 48
LLQ, 450
marking/remarking, 6869, 302304
out-of-profile, 227
PAK_priority, 153
prioritization, 679
Scavenger, 49
Scavenger-class QoS strategy, 294
unpoliced classes, 109
vertical separation of, 107
violating, 107
worm mitigation in Scavenger class, 5153
Transactional Data, 45
translating Layer 2 and Layer 3 packet markings, 90
802.1Q/p to and from DSCP, 9293
DHCP to Frame Relay DE bit, 94
IP precedence to ATM/Frame Relay PVCs,
9596
table map feature, 98
transmit queuing (Catalyst 6500), 392
transmit ring (Tx-ring), 136
troubleshooting
class naming, 520
DoS attacks (campus networks), 292294
worms (campus networks), 292294
trust boundaries
access-edge, 302
Conditionally Trusted Endpoint Models,
303, 307312
Trusted Endpoint Models, 302304,
314315
Untrusted Endpoint Models, 304307
defined, 302
trust states, 225

configuring trust on Catalyst 6500, 255
trust-device command, 386
trusted endpoint models, 302304, 314315
Catalyst 2970/3750, 343346
Catalyst 3550, 327
Catalyst 4500, 359
Catalyst 6500, 375
trusted endpoints, connecting, 304
tunnel DiffServ, 566
tunneling
cRTP, 180
modes (MPLS DiffServ), 566
Pipe Mode, 573-582
tx-queue command, 367
tx-ring-limit command, 490
Tx-rings (transmit rings), 136, 152
ATM, 489
tuning, 451
type of service (ToS), 47
U
UBR (unspecied bit rate), 491
UDP and TCP, 553554
underruns (jitter buffers), 14
unidirectional applications, 513515
unidirectional QoS, 677
unspecied bit rate (UBR), 491
Untrusted Endpoint Models (trust boundaries),
304307
Untrusted Multiapplication Server Model, 315318
show class-map and show policy-map
verification commands, 318
show mls masks qos verification command, 319
show mls qos interface policers verification
command, 318
Catalyst 2950, 315
Catalyst 2970/3750, 344
Catalyst 3550, 327329
Catalyst 4500, 359360
WAN aggregators
Catalyst 6500, 378379

show qos acl verication command,
380381
show qos maps verication command,
379380
show qos policer verication command,
381382
show qos statistics verication command,
382
Catalyst 2970/3750, 345
Catalyst 3550, 330331
Catalyst 4500, 360362
Catalyst 6500, 383386
uplink connections (DSL and cable), 677
upstream QoS, 271
V
variable network delay. See jitter
VBR (variable bit-rate), 673
verifying
ATM IMA group, 496
tag-switching configuration (MPLS per-VPN
TE), 600
vertical separation of trafc, 107
very-high-speed ATM links, 496497
video
QoS, 39
Interactive-Video, 39
Streaming-Video, 41
Streaming-Video, protecting, 557
surveillance systems, 303
videoconferencing
any-to-any, 548549
gateways and systems, 303
videoconferencing rate, 40
violating trafc, 107
viruses, 526
VoFR (Voice over Frame Relay), 149
voice
gateway packet marking, 7981
VVLANs, 314
733
Voice and Data WAN Edge Model, 454

Voice over Frame Relay (VoFR), 149
voice VLANs (VVLANs), 314
VoIP (Voice over IP), 33
bandwidth, 3638
Call-Signaling traffic, 38
campus networks, 291
header-compression techniques, 170
formats, 173174
Layer 2 encapsulation protocol support,
175176
RTP header compression (cRTP), 172
standards, 171
TCP header compression (cTCP), 171
LLQ, 141
over ATM, 91
over Ethernet to VoIP over a WAN, 91
over MPLS, 91
QoS
bearer trafc, 3438
Call-Signaling trafc, 38
traffic, dominating links, 449
VPNs (virtual private networks)
MPLS VPN QoS design, 547. See also MPLS
VPN QoS design
VVLANs (voice VLANs), 314
W
WAN aggregation router QoS design
case study, 505507
WAN aggregator/branch router handoff, 420422
WAN aggregators, 447, 548
link speeds, 452
PAK_priority, 452
required QoS policies, 448
serialization, 450
software queuing, 448449
Tx-ring tuning, 451
734
WAN edge classication and provisioning models

High Link-Speed QoS Class Models, 459
Distributed-Platform/Consistent QoS
Behavior QoS Baseline Model,
465466
Eight-Class Model, 460-462
Slow/Medium Link-Speed QoS Class Models,
454
Five-Class Model, 456459
Three-Class (Voice and Data) Model,
454-456
WAN edge link-specic QoS design
ATM
very-high-speed links, 496497
ATM-FR SIW, 497501
Frame Relay, 478
Bc, 479
Be, 480
CIR, 479
distributed platform links, 486487
ISDN
CallManager CAC limitations, 503
MLP packet reordering, 502
variable bandwidth, 501
voice and data on multiple ISDN B
channels, 503504
leased lines, 467
high-speed, 472478
medium-speed, 471
slow-speed, 467470
WAN edge QoS design, 514515
WANs, 269, 548
link fragmentation and interleaving, 181183
fragment sizes, 183184
Frame Relay fragmentation, 185, 188189

IPSec prefragmentation, 190
Multilink PPP Link Fragmentation and
Interleaving (MLP LFI), 183185
low link speeds, 450
routers roles in, 447
Weighted Random Early Detection. See WRED
WFQ, 137
wireless access points, 304
wireless IP phones, 304
WLANs (wireless LANs)
basic service set information element, 278
QoS, 270
worms, 50
campus network mitigation strategies, 292294
CodeRedv2, 527528
compared to viruses, 526
mitigation in Scavenger class, 5154
mitigation principles, 6162
NIMDA, 529
policing, 534535
preparing for future worms, 533534
RPC DCOM/W32/MS Blaster, 531532
Sasser, 532533
SQL Slammer, 530
WRED (Weighted Random Early Detection),
159164
Catalyst 3550, 340
ECN, 165
enabling on the Best-Effort class, 457
thresholds, 241
WRED-drop thresholds (Catalyst 6500), 262
wrr-queue bandwidth command, 322
wrr-queue cos map command, 240
wrr-queue dscp-map interface conguration
command, 339
wrr-queue queue-limit command, 240
wrr-queue queue-limit interface command, 338
Cisco Press
Your first-step to
networking starts here
Are you new to the world of networking? Whether you are taking your first
networking class or simply need a better understanding of a specific technology
to choose your next networking class, Cisco Press First-Step books are right
for you.
No experience required
Includes clear and easily understood explanations
Makes learning easy
Check out each of these First-Step books that cover key networking topics
Computer Networking
First-Step
ISBN: 1-58720-101-1
Network Security
First-Step
ISBN: 1-58720-099-6
TCP/IP First-Step
ISBN: 1-58720-108-9
October 2004
LAN Switching
First-Step
ISBN: 1-58720-100-3
Routing First-Step
ISBN: 1-58720-122-4
September 2004
Wireless Networks
First-Step
ISBN: 1-58720-111-9
Visit www.ciscopress.com/firststep to learn more.
Whats your next step?

Eager to dig deeper into networking technology? Cisco Press has
the books that will help you move to the next level. Learn more at
www.ciscopress.com/series.
ciscopress.com
Learning begins with a first step.
Cisco Press
NETWORK BUSINESS SERIES

JUSTIFY YOUR NETWORK INVESTMENT
Network Business books deliver:
A clear and approachable writing styleno in-depth technical knowledge required
Technology overviews that promote informed decision making
Implementation scenarios that outline various business models
ROI and TCO metrics that assist with complex technology decisions
Interviews with industry leaders that provide real-world insights on technology decisions
Detailed case studies that showcase relevant technologies
Look for Network Business titles at your

favorite bookseller
The Case for Virtual Business Processes
Young / Jude ISBN: 1-58720-087-2
IP Telephony Unveiled
Brown ISBN: 1-58720-075-9
Planet Broadband
Yassini ISBN: 1-58720-090-2
Power Up Your Small-Medium Business
Aber ISBN: 1-58705-135-4
The Road to IP Telephony
Carhee ISBN: 1-58720-088-0
Taking Charge of Your VoIP Project
Walker / Hicks ISBN: 1-58720-092-9
The Business Case for E-Learning
Kelly / Nanjiani ISBN: 1-58720-086-4 Coming Soon
Network Business Series.
Justify Your Network Investment.
Visit www.ciscopress.com/series for details about the Network Business series and a complete list
of titles.
Cisco Press
FUNDAMENTALS SERIES
ESSENTIAL EXPLANATIONS AND SOLUTIONS
When you need an authoritative introduction
to a key networking topic, reach for a Cisco
Press Fundamentals book. Learn about network
topologies, deployment concepts, protocols,
and management techniques and master
essential networking concepts and solutions.
Look for Fundamentals titles at your favorite bookseller

802.11 Wireless LAN Fundamentals
ISBN: 1-58705-077-3
IP Addressing Fundamentals
ISBN: 1-58705-067-6
Cisco CallManager Fundamentals:

A Cisco AVVID Solution
ISBN: 1-58705-008-0
IP Routing Fundamentals
ISBN: 1-57870-071-X
Data Center Fundamentals

ISBN: 1-58705-023-4
Visit www.ciscopress.com/series for details about

the Fundamentals series and a complete list of
titles.
Voice over IP Fundamentals

ISBN: 1-57870-168-6
Learning is serious business.

Invest wisely.
Cisco Press
CISCO CERTIFICATION SELF-STUDY

#1 BEST-SELLING TITLES FROM CCNA TO CCIE
Look for Cisco Press Certification Self-Study resources at

your favorite bookseller
Learn the test
topics with
Self-Study
Guides
Gain hands-on
experience with
Practical Studies
books
Prepare for the

exam with Exam
Certification
Guides
Practice testing
skills and build
confidence with
Flash Cards and
Exam Practice
Packs
Visit www.ciscopress.com/series to learn more

about the Certification Self-Study product family
and associated series.

Invest wisely.
Cisco Press
NETWORKING TECHNOLOGY GUIDES

MASTER THE NETWORK
Turn to Networking Technology Guides whenever you need in-depth knowledge
of complex networking technologies. Written by leading networking authorities,
these guides offer theoretical and practical knowledge for real-world networking
applications and solutions.
Look for Networking Technology Guides

at your favorite bookseller
Cisco Access Control Security:
AAA Administration Services
ISBN: 1-58705-124-9
Cisco CallManager Best Practices:
A Cisco AVVID Solution
ISBN: 1-58705-139-7
Designing Network Security,
Second Edition
ISBN: 1-58705-117-6
Network Security Architectures
ISBN: 1-58705-115-X
Optical Network Design
and Implementation
ISBN: 1-58705-105-2
Top-Down Network Design, Second Edition
ISBN: 1-58705-152-4
Troubleshooting Virtual Private Networks
ISBN: 1-58705-104-4
Visit www.ciscopress.com/series for details about

Networking Technology Guides and a complete list
of titles.

Invest wisely.
SEARCH THOUSANDS
OF BOOKS FROM
LEADING PUBLISHERS
Safari Bookshelf is a searchable electronic reference library for IT
professionals that features more than 2,000 titles from technical
publishers, including Cisco Press.
With Safari Bookshelf you can
Search the full text of thousands of technical books, including more than 70 Cisco Press
titles from authors such as Wendell Odom, Jeff Doyle, Bill Parkhurst, Sam Halabi, and
Karl Solie.
Read the books on My Bookshelf from cover to cover, or just flip to the information
you need.
Browse books by category to research any technical topic.
Download chapters for printing and viewing offline.
With a customized library, youll have access to your books when and where you need
themand all you need is a user name and password.
TRY SAFARI BOOKSHELF FREE FOR 14 DAYS!

You can sign up to get a 10-slot Bookshelf free for the first 14 days.
Visit http://safari.ciscopress.com to register.
Cisco Press
3 STEPS TO LEARNING
STEP 1
STEP 2
STEP 3
First-Step
Fundamentals
Networking
Technology Guides
STEP 1
First-StepBenefit from easy-to-grasp explanations.

No experience required!
STEP 2
FundamentalsUnderstand the purpose, application,

and management of technology.
STEP 3
Networking Technology GuidesGain the knowledge

to master the challenge of the network.
NETWORK BUSINESS SERIES

The Network Business series helps professionals tackle the
business issues surrounding the network. Whether you are a
seasoned IT professional or a business manager with minimal
technical expertise, this series will help you understand the
business case for technologies.
Justify Your Network Investment.
Look for Cisco Press titles at your favorite bookseller today.

Visit www.ciscopress.com/series for details on each of these book series.
Cisco Press
SAVE UP TO 25%
Become a member and save at ciscopress.com!
Complete a User Profile at ciscopress.com today to become a member and

benefit from discounts of up to 25% on every purchase at ciscopress.com,
as well as a more customized user experience. You can also sign up to get
your first 30 days FREE on InformIT Safari Bookshelf and preview Cisco Press
content. With Safari Bookshelf, you can access Cisco Press books online
and build your own customized, searchable electronic reference library.
And dont forget to subscribe to the monthly Cisco Press newsletter.
Visit www.ciscopress.com/register to sign up and start saving today!
The profile information we collect is used in aggregate to provide us with better insight into your technology
interests and to create a better user experience for you. You must be logged into ciscopress.com to receive
your discount. Discount is on Cisco Press products only; shipping and handling are not included.

Invest wisely.

End To End QoS Network Design

Uploaded by

Copyright:

Available Formats

End To End QoS Network Design

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

End To End QoS Network Design

Uploaded by

Copyright:

Available Formats

End-to-End QoS Network Design

Tim Szigeti, CCIE No. 9794, and Christina Hattingh

End-to-End QoS Network Design

Warning and Disclaimer

Corporate and Government Sales

About the Authors

About the Technical Editors

QoS Design Overview

Classication and Marking Tools

Policing and Shaping Tools

Call Admission Control (CAC)

Catalyst QoS Tools

WLAN QoS Tools

LAN QoS Design

Campus QoS Design

WAN QoS Design

WAN Aggregator QoS Design

Branch Router QoS Design

VPN QoS Design

MPLS VPN QoS Design

IPSec VPN QoS Design

QoS At-A-Glance Summaries

Chapter 1 Introduction to QoS

A Brief Historical Perspective

User Network Expectations 9

Chapter 2 QoS Design Overview

QoS Requirements of VoIP 33

QoS Requirements of Data 42

Chapter 3 Classification and Marking Tools

Token Bucket Algorithms 105

Chapter 5 Congestion-Management Tools

Understanding Scheduling and Queuing

Legacy Layer 3 Queuing Mechanisms 136

Further Reading 154

Weighted Random Early Detection

DSCP-Based Weighted Random Early Detection

Further Reading 166

Chapter 7 Link-Specific Tools

Header-Compression Techniques 170

Further Reading 191

RSVP Overview 196

Endpoints and Proxies 201

Further Reading 201

Chapter 9 Call Admission Control (CAC)

CAC Overview 205

CAC Tool Categories 207

Further Reading 218

Generic Catalyst QoS Models 224

Catalyst 2950 231

Catalyst 2970, 3650, and 3750 242

Catalyst 6500 252

Chapter 11 WLAN QoS Tools

QoS for Wireless LANs Versus QoS on Wired LANs 270

QoS Operation on Cisco APs 280

LAN QoS Design

Chapter 12 Campus QoS Design

DoS/Worm-Mitigation Strategies 292

Access-Edge Trust Models 302

Catalyst 2950: Conditionally Trusted IP Phone + PC: Advanced Model