BRKSPG-2904 Miami PDF

ASR-9000/IOS-XR hardware Architecture,
QOS, EVC, IOS-XR Configuration and

Troubleshooting
BRKSPG-2904
Aleksandar Vidakovic CCIE #5697 Technical Leader

High-End Routing and Optical group -- ASR9000
Agenda
Introduction
ASR9000 operation and capabilities
Troubleshooting ASR 9000 Forwarding
Inject/punt path architecture and troubleshooting
Unicast forwarding architecture and troubleshooting
Multicast forwarding architecture and troubleshooting
QOS architecture and troubleshooting
Scaled ACLs
IOS-XR differences to legacy IOS
Mapping IOS to XR configurations (eg EVC infrastructure)
Summary
BRKSPG-2904 2013 Cisco and/or its affiliates. All rights reserved. Cisco Public 3
ASR 9K Chassis Overview
ASR 9001 ASR 9922

ASR 9904 ASR 9006 ASR 9010 ASR 9912
(Ironman) (Megatron)
Max Capacity 770G/slot 440G/slot 440G/slot 550G/slot 1.2T/slot

120Gbps
(bi-dir) 2 I/O slots 4 I/O slots 8 I/O slots 10 I/O slots 20 I/O slot
Size 2RU 6RU 10RU 21RU 30RU 44RU
Air Flow Side to side Side to side Side to back Front to back Front to back Front to back
ASR 9000 Power System
Power Switch:
PEM Insertion from the Front V1 in the back
PEM1 V2 in the front
PEM1
PEM2 PEM2
PEM3 PEM3
PEM4
V1 Power
V2 Power
Power Feed Cabling from the Back
M3
M2 M1 M0 PWR A-, M3
V2 AC power
RTN A+, M3 PWR B-, M3 V2 DC power
RTN B+, M3
Few words about power
For DC Feed A & B loadshare
You should see ~50% distribution
Under high load conditions, all
modules should provide almost
equal power to the bus
In Low load conditions this may be
slightly off
Picture shows v1 power trays (3
per shelf). v2 has 4 modules per
shelf, same hardware, different
formfactor.
Efficiency near 98%, use 90% for
calculation
Each DC feed needs breaker for
max amp: (2.1K/48V)/0.9 ~= 50A
All modules feed the bus, RSPs
booted first with Fans, LCs next
starting slot 0 until avail power is
gone
Command admin show env power
Example output
RP/0/RSP0/CPU0:WEST-PE-ASR9K-2#admin show platform
Tue Jan 28 09:23:00.442 EST
Node Type State Config State
-----------------------------------------------------------------------------
0/RSP0/CPU0 A9K-RSP-8G(Active) IOS XR RUN PWR,NSHUT,MON
0/RSP1/CPU0 A9K-RSP-8G(Standby) IN-RESET PWR,NSHUT,MON
0/FT0/SP FAN TRAY READY
0/FT1/SP FAN TRAY READY
0/0/CPU0 A9K-40GE-E IOS XR RUN PWR,NSHUT,MON
0/1/CPU0 A9K-MOD160-TR IOS XR RUN PWR,NSHUT,MON
0/1/1 A9K-MPA-20X1GE OK PWR,NSHUT,MON
0/6/CPU0 A9K-SIP-700 IN-RESET PWR,NSHUT,MON
0/PM0/SP A9K-3KW-AC READY PWR,NSHUT,MON
0/PM2/SP A9K-3KW-AC FAILED PWR,NSHUT,MON
Bold sections indicate what would be omitted if show platform was ran in exec mode
Example output
RP/0/RSP0/CPU0:WEST-PE-ASR9K-2#admin show environment power-supply
Tue Jan 28 09:17:11.021 EST
R/S/I Modules Capacity Status
(W)
0/PM0/*
host PM 3000 Ok Module status
0/PM1/*
host PM 0 Powered (No output)
0/PM2/*
host PM 0 Powered (No output)
R/S/I Power Supply Voltage Current

(W) (V) (A) Actual draw
0/PM0/* 573.3 54.6 10.5
0/PM1/* 0.0 0.0 0.0
0/PM2/* 0.0 0.0 0.0
--------------
Total: 573.3
Example output
R/S/I Power Draw Voltage Current
(W) (V) (A)
0/FT0/* 300 **
0/FT1/* 300 **
--------------
Total: 600.0

(W) (V) (A)
0/RSP0/* 205 **
--------------
Total: 205.0

(W) (V) (A)
0/0/* 320 ** Actual draw
0/1/* 486.0 54.0 9.0
-------------- (only Typhoon, RSP440)
Total: 806.0
** Indicates worst case power draw
Example output
Power Budget Summary for Rack 0
-------------------------------- Software budget table (based on defined
temperature profile)
Power Shelves Type: AC
Power Budget Strict Mode: Disabled Hard coded, used by power manager to
determine cards to boot
Total Power Capacity: 3000W
Usable Power Capacity: 3000W
Supply Failure Protected Capacity: 0W
Feed Failure Protected Capacity: 0W
Worst Case Power Used: 1920W
Slot Max Watts

---- ---------
0/0/CPU0 320
0/1/CPU0 590 Standby RSP
0/RSP0/CPU0 205 is allocated for
0/RSP1/CPU0 205 by default,
0/FT0/SP 300 even when not
0/FT1/SP 300 present
Worst Case Power Available: 1080W
Supply Protected Capacity Available: Not Protected
Feed Protected Capacity Available: Not Protected
Packet Flow Overview
Same as existing system: Two-stage IOS-XR packet forwarding
Uniform packet flow: All packet go through central fabric on the RP
3x 10G
3x10GE
SFP + Typhoon
3x10GE
3x 10G FIA
SFP + Typhoon
FIA Ingress 100G
3x 10G Typhoon
3x10GE 100GE
ASIC
Switch Fabric
SFP + Typhoon
MAC/PHY
3x 10G FIA Switch
ASIC
Switch Fabric
3x10GE Egress 100G
SFP + Typhoon Fabric FIA Typhoon
3x 10G
3x10GE
SFP + Typhoon
3x 10G FIA FIA
Ingress
3x10GE Typhoon 100G
SFP + Typhoon
100GE
3x 10G
3x10GE MAC/PHY
SFP + Typhoon
Egress
3x10GE
3x 10G FIA Switch FIA Typhoon 100G
SFP + Typhoon Fabric
Fabric Arbitration and Redundancy
0 packet loss guarantee during RSP failover and OIR
Physically separated from LC. Resides on RSP or on a dedicated card
(9912, 9922)
Logically separated from LC and RSP
All fabric ASICs run in active mode regardless of RSP Redundancy
status
Extra fabric bandwidth and instant fabric switch over
If the FAB has been previously initiated then even with RP in rommon
FABRIC IS ACTIVE!
Access to fabric controlled using central arbitration.
One Arbitration ASIC (Arbiter) per RSP
Both Arbiters work in parallel both answer to requests to transmit
FIAs follow active Arbiter, and switch to backup if needed
Arbiter switchover controlled by low level hardware signalling
Fabric Arbitration
RSP0
Crossbar
Fabric 5: credit return
ASIC
1: Fabric Request
Crossbar
Fabric
ASIC
Arbitration
Fabric Interface
Fabric Interface
and VOQ 2: Arbitration and VOQ
Crossbar
Fabric
3: Fabric Grant ASIC
Crossbar
4: load-balanced Fabric
ASIC
transmission across
Arbitration
fabric links
RSP1
Fabric Load Sharing Unicast
Crossbar
Fabric
ASIC
Crossbar
Fabric
ASIC
Arbitration
Fabric Interface
Fabric Interface
RSP0 and VOQ
4 3 2 1
and VOQ
Crossbar
Fabric
ASIC
Crossbar
Fabric
ASIC
Arbitration
RSP1
Unicast traffic sent across first available fabric link to destination (maximizes efficiency)
Each frame (or superframe) contains sequencing information
All destination fabric interface ASIC have re-sequencing logic
Additional re-sequencing latency is measured
BRKSPG-2904 in nanoseconds
2013 Cisco and/or its affiliates. All rights reserved. Cisco Public 34
Fabric Load Sharing Multicast
Crossbar
Fabric
ASIC
Crossbar
Fabric
ASIC
Arbitration
Fabric Interface C1 B2 A3 B1 A2 A1
Fabric Interface
and VOQ
RSP0 and VOQ
Flows exit in-order

Crossbar
Fabric
ASIC
Crossbar
Fabric
ASIC
Arbitration
RSP1
Multicast traffic hashed based on (S,G) info to maintain flow integrity

Very large set of multicast destinations preclude re-sequencing
Multicast traffic is non arbitrated sent across a different fabric plane
Fabric Super-framing Mechanism
Multiple unicast frames from/to same destinations aggregated into one super frame
Super frame is created if there are frames waiting in the queue, up to 32 frames or
when min threshold met, can be aggregated into one super frame
Super frame only apply to unicast, not multicast
Super-framing significantly improves total fabric throughput
Packet 1 No super-framing
Packet 2 Packet 1 Min reached
Packet 3 Packet 2 Packet 1 Max reached
Packet 1 Jumbo
Max Min 0 (Empty)

Super-frame Super-frame
Note that fabric counters are showing super frames not individual packets!!
(show controller fabric fia stats loc 0/X/CPU0)
Fabric packet counters
RP/0/RSP0/CPU0:A9K-2#sh contr fabric fia stat loc 0/1/CPU0 | utility egrep "\] +[1-9]|^C|FIA"
Tue Jan 28 09:33:40.032 EST
********** FIA-0 **********
Category: count-0
From Unicast Xbar[0] 12684
From Unicast Xbar[1] 12938
To Unicast Xbar[0] 12870
To Unicast Xbar[1] 12752
To Line Interface[0] 25622
From Line Interface[0] 25622
********** FIA-1 **********
Category: count-1
<>
********** FIA-2 **********
Category: count-2
<>
********** FIA-3 **********
Category: count-3
<>
Note: output is showing a system with
one RSP (ie. one fabric card)
Meaning of hard drop -x reason in
sh controllers fabric fia drops [ingress|egress]
There are four priority levels and four physical XBAR links. Now the confusion is that, fia egress
drop stats are per priority, while fia ingress drop stats are per XBAR link.
The fia egress drop stats, Tail, Hard, WRED, (offsets 0-3) represent fabric priority stats and
correspond as...
0 - high priority level 1
1 - high priority level 2
2 - low priority
3 - not used (asr9k)
The fia ingress drop stats offsets (0-3) represent XBAR link stats and correspond as...
0-1 XBAR links to RSP0 (Trident+RSP2)

2-3 XBAR links to RSP1 (Trident+RSP2)
On Typhoon cards the FIA links with 2 links to the local fabric.
The local fabric connects with 8x55G links to the RSP fabric
Fabric drop counters - ingress
RP/0/RSP0/CPU0:A9K-2#sh controllers fabric fia drops ingress location 0/1/CPU0
Tue Jan 28 09:40:35.255 EST
********** FIA-0 **********

Category: in_drop-0
From Spaui Drop-0 0 Header parsing drp 0
accpt tbl-0 0 pw to ni drp 0
ctl len-0 0 ni from pw drp 0
short pkt-0 0 sp0 crc err 8
max pkt len-0 0 sp0 bad align 0
min pkt len-0 0 sp0 bad code 2
From Spaui Drop-1 0 sp0 align fail 3
accpt tbl-1 0 sp0 prot err 0
ctl len-1 0 sp1 crc err 3
short pkt-1 0 sp1 bad align 0
max pkt len-1 0 sp1 bad code 2
min pkt len-1 0 sp1 align fail 3
Tail drp 0 sp1 prot err 0
Vqi drp 0
********** FIA-0 **********

<> BRKSPG-2904 2013 Cisco and/or its affiliates. All rights reserved. Cisco Public 39
Fabric drop counters - egress
RP/0/RSP0/CPU0:A9K-2#sh contr fabric fia drops egress location 0/1/CPU0
Tue Jan 28 09:51:03.746 EST
Priority
********** FIA-0 **********
Category: eg_drop-0
From Xbar Uc Crc-0 0 Uc dq pkt-len-crc/RO-seq/len error drp 0
From Xbar Uc Crc-1 0 Mc rf crc drp 0
From Xbar Uc Crc-2 0 Mc vl0 src0 buffer full drp 0
From Xbar Uc Crc-3 0 Mc vl1 src0 buffer full drp 0
From Xbar Uc Drp-0 0 Mc vl2 src0 buffer full drp 0
From Xbar Mc Crc-0 0 Mc vl2 src1 buffer full drp 0
From Xbar Mc Crc-1 0 Mc vl3 src1 buffer full drp 0
From Xbar Mc Crc-2 0
From Xbar Mc Crc-3 0
From Xbar Mc Drp-0 0
********** FIA-1 **********

Fabric counters on Typhoon (3-stage fabric)
RP/0/RSP0/CPU0:A9K-2#sh controllers fabric crossbar statistics instance 0 location 0/1/CPU0
Tue Jan 28 10:00:38.306 EST
Port statistics for xbar:0 port:0
==============================
Hi priority stats (unicast)
===========================
Ingress Packet Count Since Last Read : 12821
Egress Packet Count Since Last Read : 9590
Low priority stats (multicast)

===========================
Port statistics for xbar:0 port:1

==============================
Hi priority stats (unicast)
===========================
Ingress Packet Count Since Last Read : 12782
Egress Packet Count Since Last Read : 9778
Low priority stats (multicast)

===========================
<>
ASR 9001 System Architecture Overview
MPAs
2,4x10GE
20xGE
1x40GE
Typhoon FIA
ASIC
Switch Fabric
SFP+ 10GE
SFP+ 10GE
On-board 4x10
SFP+ ports Internal LC RP
SFP+ 10GE EOBC CPU CPU
SFP+ 10GE
MPAs Typhoon
2,4x10GE
20xGE
FIA
1x40GE
It has both central RP and LC CPU like big chassis

But it only have central switch fabric, no LC fabric
Maximum 120Gbps bi-directional system.
9001-S, a 60G version is available with only 1 Bay enabled, can upgrade to 120G via license
ASR 9001 Packet Flow Overview
Same as big chassis system:
Supported MPA Two-stage IOS-XR packet forwarding
2,4x10GE
20xGE
1x40GE
Typhoon FIA
ASIC
Switch Fabric
SFP+ 10GE
SFP+ 10GE
Internal LC RP
SFP+ 10GE EOBC CPU CPU
SFP+ 10GE
Typhoon
Supported MPA
FIA
2,4x10GE
20xGE
1x40GE
Troubleshooting ASR9000 Forwarding
NPU Packet Processing - Ingress
5 Stages: All packets go through

the TM regardless of
whether QOS is enabled
Queueing
Parse Search Resolve Modify Scheduling
L2/L3 header Performs QoS Processes Adds internal Queuing,

packet parsing in and ACL Search results: system headers Shaping and
TCAM lookups in ACL filtering Egress Control Scheduling
Builds keys for TCAM tables Ingress QoS Header (ECH) functions
ingress ACL, Performs L2 classification Switch Fabric
QoS and and L3 and policing Header (SFH)
forwarding lookups in Forwarding
lookups (uCode) RLDRAM (egress SFP
determined)
Performs L2
MAC learning
Where to start when there are forwarding issues
First identify interface in question with problem

Identify the mapping from interface to NPU
Show controller np ports all location 0/X/CPU0 (where x is the slot)
Show the controller NPU counters
Show controller np count npY location 0/X/CPU0 (where y is the NPU for IF)
Look for rate counters that match lost traffic rate
Lookup description for counter (see next slide)
Check FIA counters
Check fabric counters
Move to egress interface and repeat steps 2 and 3.
Port to NPU mapping
RP/0/RSP0/CPU0:A9K-BNG#show controller np ports all loc 0/0/cpU0
Node: 0/0/CPU0:
----------------------------------------------------------------
NP Bridge Fia Ports

-- ------ --- ---------------------------------------------------
0 -- 0 GigabitEthernet0/0/0/0 - GigabitEthernet0/0/0/9
1 -- 1 GigabitEthernet0/0/0/10 - GigabitEthernet0/0/0/19
2 -- 2 TenGigE0/0/1/0
3 -- 3 TenGigE0/0/1/1
Example
RP/0/RSP0/CPU0:A9K-BNG#show controller np counters np0 loc 0/0/CPU0
Node: 0/0/CPU0:
----------------------------------------------------------------
Show global stats counters for NP0, revision v2 MDF=Modify
Read 57 non-zero NP counters: TX transmit
WIRE to the
Offset Counter FrameValue Rate (pps) wire = egress
-------------------------------------------------------------------------------
16 MDF_TX_LC_CPU 22755787 6
17 MDF_TX_WIRE 1614696 0
21 MDF_TX_FABRIC 1530106 0
29 PARSE_FAB_RECEIVE_CNT 1555034 0
33 PARSE_INTR_RECEIVE_CNT 22026578 6
Packets
received from
37 PARSE_INJ_RECEIVE_CNT 335774 0 the fabric
41 PARSE_ENET_RECEIVE_CNT 2115361 1
45 PARSE_TM_LOOP_RECEIVE_CNT 17539300 5
Delta between received from Fab to TX-wire should almost be 0, if not, we

dropped packets, could be ACL, QOS, or for other reasons (eg PUNT)
Note
Some counters have an index to a port.

For instance, there is an aggregate count per NPU showing the misses
from vlan to subinterface mapping:
UIDB_TCAM_MISS_AGG_DROP
There is also a specific counter from which port index these drops came
from:
UIDB_TCAM_MISS_DROP_1
This means that the second port (starting count from zero) on that NPU
experienced that drop.
So if your show controller np ports tells us that ports X Y and Z are
connected to this NPU, and the drop index is _1, then port Y is the
culprit.
Capturing lost packets in the NPU
CLI:
monitor np counter <COUNTER_NAME> <NPU> count <N>
You can monitor any counter in the NPU on Typhoon generation line cards
Captured packets are always dropped
Exists automatically after capturing <N> packets or when timeout is reached
NPU is reset upon exit (~50ms forwarding stop)
This will be enhanced later
Packets subject to punt cannot be captured by this methodology
Use with care!
Capturing lost packets in the NPU - Example
RP/0/RSP0/CPU0:A9K-2#monitor np counter PRS_HEALTH_MON np0 count 1 location 0/1/CPU0
Tue Jan 28 10:10:18.824 EST
Warning: Every packet captured will be dropped! If you use the 'count'
option to capture multiple protocol packets, this could disrupt
protocol sessions (eg, OSPF session flap). So if capturing protocol
packets, capture only 1 at a time.
Warning: A mandatory NP reset will be done after monitor to clean up.

This will cause ~50ms traffic outage. Links will stay Up.
Proceed y/n [y] >
Monitor PRS_HEALTH_MON on NP0 ... (Ctrl-C to quit)
Tue Jan 28 10:10:22 2014 -- NP0 packet
From OAM RTC: 64 byte packet, bytes[0-3] invalid!

0000: 40 00 00 48 80 00 00 00 00 00 00 00 00 00 10 00 @..H............
0010: 00 00 00 00 00 00 00 00 04 55 aa 55 aa 55 aa 55 .........U*U*U*U
0020: aa 55 aa 55 aa 55 aa 55 aa 55 aa 55 aa 55 aa 55 *U*U*U*U*U*U*U*U
<>
0190: 00 00 00 00 00 00 00 00 ........
(count 1 of 1)
Cleanup: Confirm NP reset now (~50ms traffic outage).

Ready? [y] >
RP/0/RSP0/CPU0:A9K-2#
Troubleshooting ASR9000 Forwarding: Punt/Inject
IOS XR Control Plane
Local Packet Transport Service
packets in
LPTS App 1
Control Plane Traffic RP
good packets
User Traffic
for-us packets
LPTS Internal FIB

FIB (IFIB) App 2
RP
LC DCoPP
Dynamic Control
transit
Plane Policing bad packets
Local
packets out
Stacks LC
LPTS enables applications to reside on any or all RPs, DRPs, or LCs

Active/Standby, Distributed Applications, Local processing
IFIB forwarding is based on matching control plane flows
DCoPP is built in firewall for control plane traffic.
LPTS is transparent and automatic
IOS XR LPTS in action
LPTS is an automatic, built in firewall for
Router bgp
control plane traffic. neighbor 202.4.48.99
ttl_security
Every Control and Management packet from ! mpls ldp
the line card is rate limited in hardware to
!
protect RP and LC CPU from attacks
LC 1 IFIB TCAM HW Entries
Local port Remote port Rate Priority
Any ICMP ANY ANY 1000 low
Socket
LPTS
any 179 any any 100 medium BGP
any 179 202.4.48.99 any 1000 medium ttl
202.4.48.1 179 202.4.48.99 2223 10000 medium 255
200.200.0.2 13232 200.200.0.1 646 100 medium
LDP
SSH
LC 2 IFIB TCAM HW Entries

TCP Handshake
Tightening LPTS
If you can use only p2p OSPF network type

flow OSPF-uc-known rate 0
flow OSPF-uc-default rate 0
Note that OSPF p2p network type is the recommended setting even
on Ethernet interfaces unless you have multiple routers on the same
segment.
Do we really need BGP, LDP-TCP, MSDP, default for unconfigured
sessions
Flow BGP-default rate 0
Flow LDP-tcp-default rate 0
Flow MSDP-default rate 0
Further investigation needed for the following
flow udp-default rate 0
flow tcp-default rate 0
flow raw-default rate 0
DCN
Management Plane Protection (MPP) IP/MPLS
In-band MPP
I/F 1 is configured as MPP I/F 1
in-band interface. I/F 1 is
also part of global
routing/forwarding. LPTS
Management traffic to RP
from all non-MPP R I
interfaces is dropped (I/F P RP /
2 and I/F 3). CPU F
E
RP Eth/Console/Aux t 3
continues to operate as h
dedicated out-of-band.
LPTS still continues to
provide rate limiting
irrespective of MPP. I/F 2
Troubleshooting MPP -- LPTS
control-plane RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh lpts bindings brief | i "any,23 "
management-plane 0/RSP0/CPU0 TCP LR IPV4 TCP default Mg0/RSP0/CPU0/0 any,23 any
inband 0/RSP0/CPU0 TCP LR IPV4 TCP default Mg0/RSP0/CPU0/1 any,23 any
interface Loopback0 0/RSP0/CPU0 TCP LR IPV4 TCP default Te0/4/0/1 any,23 any
allow SNMP 0/RSP0/CPU0 TCP LR IPV4 TCP default Gi0/7/0/7 any,23 40.0.60.2
! RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#
interface TenGigE0/4/0/1 RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh lpts bindings brief | i "any,161 "
allow Telnet 0/RSP0/CPU0 UDP LR IPV4 UDP default Mg0/RSP0/CPU0/0 any,161 any
! 0/RSP0/CPU0 UDP LR IPV4 UDP default Mg0/RSP0/CPU0/1 any,161 any
interface 0/RSP0/CPU0 UDP LR IPV4 UDP default Lo0 any,161 any
GigabitEthernet0/7/0/7 0/RSP0/CPU0 UDP LR IPV6 UDP default Mg0/RSP0/CPU0/0 any,161 any
allow Telnet peer 0/RSP0/CPU0 UDP LR IPV6 UDP default Mg0/RSP0/CPU0/1 any,161 any
address ipv4 40.0.60.2 0/RSP0/CPU0 UDP LR IPV6 UDP default Lo0 any,161 any
! RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#
! RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh lpts ifib brief | i "any,23 "
! TCP4 default TCP Gi0/7/0/7 0/RSP0/CPU0 any,23 40.0.60.2
! TCP4 default TCP Mg0/RSP0/CPU0/0 0/RSP0/CPU0 any,23 any
! TCP4 default TCP Mg0/RSP0/CPU0/1 0/RSP0/CPU0 any,23 any
TCP4 default TCP Te0/4/0/1 0/RSP0/CPU0 any,23 any
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh lpts pifib hardware entry brief location 0/4/CPU0 | i any,23

19 IPV4 default TCP Te0/4/0/1 LU(30) any,23 any,any
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh lpts pifib hardware entry brief location 0/7/CPU0 | i any,23
35 IPV4 default TCP Gi0/7/0/7 LU(30) any,23 40.0.60.2,any
BRKSPG-2904 2013 Cisco and/or its affiliates. All rights reserved. Cisco Public
Verifying LPTS policer values
RP/0/RP0/CPU0:CRS1-4#show lpts pifib hardware police location 0/7/CPU0
-------------------------------------------------------------
Node 0/7/CPU0: lpts pifib hardware police
------------------------------------------------------------- flow fragment rate 0
Burst = 100ms for all flow types flow bgp default rate 0
-------------------------------------------------------------
FlowType Policer Type Cur. Rate Def. Rate Accepted Dropped
---------------------- ------- ------- ---------- ---------- ---------- ----------
unconfigured-default 100 Static 500 500 0 0
Fragment 106 Global 0 1000 0 0
OSPF-mc-known 107 Static 20000 20000 0 0
OSPF-mc-default 111 Static 5000 5000 0 0
OSPF-uc-known 161 Static 5000 5000 0 0
OSPF-uc-default 162 Static 1000 1000 0 0
BGP-known 113 Static 25000 25000 18263 0
BGP-cfg-peer 114 Static 10000 10000 6 0
BGP-default 115 Global 0 10000 0 2
PIM-mcast 116 Static 23000 23000 19186 0
PIM-ucast 117 Static 10000 10000 0 0
IGMP 118 Static 3500 3500 9441 0
ICMP-local 119 Static 2500 2500 1020 0
ICMP-app 120 Static 2500 2500 0 0
na 164 Static 2500 2500 72 0
LDP-TCP-cfg-peer 152 Static 10000 10000 0 0
LDP-TCP-default 154 Static 10000 10000 0 0
cutBRKSPG-2904 2013 Cisco and/or its affiliates. All rights reserved. Cisco Public 58
10000 ft view of Receive packet path
Ingress LC RP
CPU
LC-CPU
F
4a 4b A Egress LC
B
PIFIB
ucode
(TCAM, dynamic) R For-us traffic such as
L2, BFD, ARP
3b 3c
NP 3a I For-us traffic processed
by LPTS mgmt, L3
1 2 ucode C control plane, ICMP
Transit traffic
Legend to previous slide
2. Ingress NPU in the LC will perform packet lookup using the HW FIB to determine how to switch the packet.
3a. If the incoming packet is part of transit traffic, they will be switched by the LC HW and sent to the egress LC through the fabric
3a. If the incoming packet is of L2 type such as CDP, ARP, LACP PDU, BFD, CFM/OAM etc FIB will punt them to LC CPU for further processing.
Also transit traffic to be forwarded, but frag required Packets with DF bit set packets, IP options packet, packets with RA, transit traffic dropped by
ACL etc will be punted to LC CPU
3c. If FIB lookup determines that this is a for-us control/management plane packet, then further lookup has to be performed on the pre-IFIB
table in the HW/TCAM to match it against a flow entry, perform policing on the packet stream, and ascertain the node/element and application to
deliver
4a. For some of the for-us control packets, which needs to be delivered locally, requiring special handling such as ICMP echo, TTL expired
packets, HW Pre-IFIB look-up will punt the packets to LC CPU
4b. for us control packets destined to the RP.
6. IFIB slice lookup on a local node will provide transport and the associated application/server processes the packet needs to be delivered
Detailed packet path of for-us packets
Trident vs Typhoon LC details
Item Trident LC Typhoon LC
CPU Port TSEC (2x1G): DPAA (1x10G) RO LC: spp_ui> ioctrl mib (clear on Read)
TSEC2 / TSEC3
XMEN LC: show controllers dpaa tsec port 9 location <>
Punt Switch 10 port / 16 port 24x1G + 2x10G Show controllers punt-switch mac-stats <> location <>
(1G) Port24: 10G DPAA
Port7: TSEC2 Port10: 1G DPAA
Port8: TSEC3 Port [0 (N-1)]: NP [0
Port[0..(N-1)]: NP (N-1)]
[0 (N-1)]
(exception 8 NP LC)
NP Trident Typhoon Show controllers np ports all location <>

Show controllers np fabric-counters <rx | tx> <np> location <>
Show controllers np counters <np> location <>
FIA Octopus Skytrain Show controllers fabric fia statistics location <>
Bridge Punt N.A (integrated into Show controllers fabric bridge stats loc <>
Skytrain
Fabric (XBAR) N.A Sacramento Show controllers fabric Crossbar instance <> statistics location <>
LPTS recap
Local Packet Transport System
Pre-IFIB packet processing (for-us packets)
Control plane for Control packets
L3 applications on RSP responsible for triggering / installation of the LPTS entries
LPTS entries are installed in software (on the local CPU) and in hardware (TCAM)
3 categories
Default entries (TCAM) : L3
Dynamic entries (TCAM) : L3
Static entries (NP SRAM) : L2 / internal interest
show lpts pifib hardware entry type <ipv4 | ipv6> brief location <LC>
show lpts pifib hardware entry type <ipv4 | ipv6> statistics location <LC>
show prm server tcam .
show lpts pifib hardware static-police location <LC>
Displays the Static punt table stats. Ie. Exception packet punts
(PRM is platform resource manager, the entity that controls the hw programming between CPU nad NPU+its attached asics/memory)
SPP Rx on RSP
show spp interface location RSP
show spp node-counters location RSP
show spp node location RSP
show spp sid stats location RSP
Updated by the classification node based on SID lookup
show spp client location RSP
show spp buffer location RSP
show spp graph location RSP
run
spp_ui
help trace.
trace. Look for classify or punt or drop nodes
Note that trace filter node tsec3/rx is not allowed as Packet capture at this node is not possible currently; tsec3/classify is
the very 1st trace-able node in the Rx direction
Netio Rx on RSP
show netio idb FINT location RSP
show netio drops location RSP
debug netio drivers location RSP [filter packets coming in from fabric]
very chatty!!!
debug lpts packet [for debugging packets of type PKT_LPTS]; use
drops, detail, errors, etc.
Punt FPGA (on RSP)
show controllers fabric fia bridge.. on RSP
Not all CLI sub-options applicable to RSP
Use the following sub-options
ddr-status [look for SYNC status]
stats
flow-control
clear controller fabric fia loc RSP
Clears all of Punt FPGA, FIA counters on RSP
admin mode: show hw-module fpd location RSP
Look for any mismatches and need for up-grade/down-grade.
Most likely issue of drops in hardware is due to FPD change requirements.
RSP2
show controllers fabric fia <drops | errors> <ingress | egress> loc <RSP>
show controllers fabric fia bridge ddr-status show controllers fabric fia link-status loc <RSP>
loc <RSP>
show controllers fabric show controllers fabric fia show controllers fabric crossbar
run spp_ui instance <> statistics location
fia bridge stats location stats location <RSP>
spp_ui> ioctrl mib <RSP> <RSP>
8641D Punt
FIA Fabric
CPU TSEC3 FPGA DDR
RSP440
show controllers fabric fia <drops | errors> <ingress | egress> loc <RSP>
show controllers fabric fia bridge ddr-
show controllers fabric fia link-status loc <RSP>
status loc <RSP>
show controllers fabric show controllers fabric

show controllers dmac show controllers fabric
fia stats location crossbar instance <>
queue 0 statistics fia bridge stats
<RSP> statistics location <RSP>
location <RSP> location <RSP>
Jasper Dao Skytrain Sacramento

Forest CPU 10G FPGA (FIA) (XBAR)
DDR
DMAC
Trident LC
Show controllers np ports all loc <LC> show controllers fabric fia bridge
Show controllers np counters <> location ddr-status loc <LC>
<LC> Show controllers fabric fia bridge
Show controllers np fabric-counters <rx | show controllers fabric
flow-control loc <LC>
tx> <np> loc <LC> fia link-status loc <LC>
show controllers fabric fia bridge
Show controllers np punt-path-counters Show controllers fabric
sync-status loc <LC>
<rx | tx> HOST-SGMII-0 <np> loc <LC> fia <drops | errors> <ing
Show spp sid stats Show lpts pifib hardware entry type | egr > loc <LC>
loc <> <ipv4 | ipv6> statis loc <LC>
Show spp node-
counters loc <>
Show spp interface Show controllers punt- Show controllers Show controllers
loc <> switch port-status loc fabric fia bridge stats fabric fia stats location
Spp_ui > ioctrl mib <LC> location <LC> <LC>
show controllers punt-
switch mac-stats <>
location <LC>
8641D TSEC2 Punt Octopus

NP Bridge
CPU TSEC3
Switch (FIA)
Show controllers
TM
Show qoshal punt-queue np <> loc <LC> MAC <interface> stats
show qoshal default-queue port <> loc <LC>
Typhoon LC
Show controllers np ports all loc <LC>

Show controllers np counters <> location <LC>
Show controllers np fabric-counters <rx | tx>
<np> loc <LC> show controllers fabric fia
Show controllers np punt-path-counters <rx | tx> link-status loc <LC>
Show spp sid stats loc <> HOST-SGMII-0 <np> loc <LC> Show controllers fabric fia
Show spp node-counters Show lpts pifib hardware entry type <ipv4 | ipv6> <drops | errors> <ing | egr
loc <> statis loc <LC> > loc <LC>
Show spp interface loc <>
show controllers dpaa
tsec port 9 location Show controllers punt-switch
Show controllers fabric
0/3/CPU0 port-status loc <LC>
Show controllers fabric Crossbar instance <>
show controllers punt-switch
fia stats location <LC> statistics location <LC>
mac-stats <> location <LC>
P4040 1G DPAA Punt Skytrain Sacramento

NP (FIA)
CPU 10G DPAA
Switch (XBAR)
Show controllers
TM
Show qoshal punt-queue np <> loc <LC> MAC <interface> stats
show qoshal default-queue port <> loc <LC>
Troubleshooting ASR9000 Forwarding: Unicast
L3 IPv4 Control Plane Architecture
BGP OSPF
LDP RSVP-TE
Static
ISIS EIGRP
LSD RIB RSP CPU
ARP
SW FIB FIB Adjacency
AIB
LC NPU
AIB: Adjacency Information Base
LC CPU RIB: Routing Information Base
FIB: Forwarding Information Base
LSD: Label Switch Database
L3 IPv4 Control Plane Architecture 2 stage lookup

BGP OSPF RIB: Routing Information Base
LDP RSVP-TE
Static FIB: Forwarding Information Base
ISIS EIGRP LSD: Label Switch Database
LSD RIB RSP CPU
ARP SW FIB HW FIB

HW FIB SW FIB
LC NPU
LC NPU
AIB
Ingress LC LC CPU LC CPU Egress LC
L3 IPv4 Control Plane Architecture 2 stage lookup

BGP OSPF RIB: Routing Information Base
LDP RSVP-TE
Static FIB: Forwarding Information Base
ISIS EIGRP LSD: Label Switch Database
Ingress Lookup
LSD RIB RSP CPU
Egress Lookup
ARP SW FIB HW FIB

HW FIB SW FIB
LC NPU
LC NPU
AIB
Ingress LC LC CPU LC CPU Egress LC
L3FIB
L3 NPU IPv4 FIB Architecture
Adjacency
32ways pointer(s) 1xLDI
Non-Recursive NR NR NR
Adj OIF
Prefix Leaf LDI LDI LDI
LAG (64
Adj
members)
OIF
8 or 32 ways
Recursive R R R R
Prefix Leaf LDI LDI LDI LDI Protected TE
Adj
LAG OIF
Adjacency
Non-Recursive pointer
LDI
Backup TE Adj OIF

Recursive Recursive
Prefix Leaf LDI
Show commands
BGP OSPF
LDP RSVP-TE
Static
ISIS EIGRP
LSD RIB RSP CPU
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh route 10.1.2.0/24
Routing entry for 10.1.2.0/24

KnownARP
via "static", distance 1, metric 0
Installed Jan 29 05:54:26.182 for 00:00:04
Routing Descriptor Blocks
SW FIB FIB Adjacency
40.0.3.1,
AIB via GigabitEthernet0/7/0/2
Route metric is 0 LC NPU
No advertising protos.
LC CPU
Show commands
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh cef 10.1.2.0/24 location 0/4/CPU0
10.1.2.0/24, version 256779, internal 0x4004001 (ptr 0xa3d59b84) [1], 0x0 (0xa3610aa0),
0x440 (0xa4fb4d50)
Updated Jan 29 05:54:26.191 BGP OSPF
remote adjacencyLDP RSVP-TE
to GigabitEthernet0/7/0/2 Static
Prefix Len 24, traffic index 0, precedence routine (0), priority 3
ISIS EIGRP
via 40.0.3.1, GigabitEthernet0/7/0/2, 8 dependencies, weight 0, class 0 [flags 0x0]
path-idx 0 [0xa51a43b4 0xa5a5a5b8]
next hop 40.0.3.1
remote adjacency
local label 17012
LSD labels imposed {ImplNull}
RIB
RSP CPU
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh cef 10.1.2.0/24 location 0/7/CPU0
10.1.2.0/24, version 256779, internal 0x4004001 (ptr 0xa4d3ee6c) [1], 0x0 (0xa35bdc80),
0x440 (0xa5b1bd50)
Updated Jan 29 05:54:26.192
local adjacency 40.0.3.1
ARP GigabitEthernet0/7/0/2, 9 dependencies, weight 0, class 0 [flags 0x0]
via 40.0.3.1,
path-idx 0 [0xa59683c8 0xa5a30268]
next hop 40.0.3.1 SW FIB FIB Adjacency
local adjacency
AIB
local label 17012 labels imposed {ImplNull} LC NPU
Entry missing or not what you have expected?LC CPU

sh adjacency trace all location
sh cef trace location
Show commands
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh adjacency summary location 0/7/CPU0
Adjacency table
LDP (version 459) has 95 adjacencies: BGP
RSVP-TE
OSPF
49 complete adjacencies Static
ISIS EIGRP
46 incomplete adjacencies
0 deleted adjacencies in quarantine list
21 adjacencies of type IPv4
21 completeLSD
adjacencies of type IPv4 RIB RSP CPU
0 incomplete adjacencies of type IPv4
0 deleted adjacencies of type IPv4 in quarantine list
0 interface adjacencies of type IPv4
12 multicast adjacencies of type IPv4
3 adjacencies of type IPv6
3 complete adjacencies of type IPv6
ARP
0 incomplete adjacencies of type IPv6
0 deleted adjacencies FIB list
of type IPv6 in quarantine
SW FIB Adjacency
0AIB
interface adjacencies of type IPv6
1 multicast adjacency of type IPv6 LC NPU
LC CPU
Show commands
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh cef 10.1.2.0/24 hardware ingress location 0/4/CPU0
10.1.2.0/24, version 256779, internal 0x4004001 (ptr 0xa3d59b84) [1], 0x0 (0xa3610aa0), 0x440 (0xa4fb4d50)
Updated Jan 29 05:54:26.191
remote adjacency to GigabitEthernet0/7/0/2
path-idx 0 [0xa51a43b4 0xa5a5a5b8]
next hop 40.0.3.1
remote adjacency
local label 17012 labels imposed {ImplNull}
LEAF - HAL pd context :
sub-type : IPV4, ecd_marked:0, has_collapsed_ldi:0, collapse_bwalk_required:0, ecdv2_marked:0
Leaf H/W Result: More info on the
<> leaf (recursion?,
fast_switch_entry
rx_adj_field: 0x5300 (LE) Where to send the BGP policy
egress_ifhandle: 0x12000140 (LE) accounting, etc.)
<> packet
nrLDI eng ctx:
flags: 0x541, proto: 2, npaths: 0, nbuckets: 1 ldi_tbl_idx: 0x0, ecd_ref_cft: 0
RX H/W Result on NP:0 [Adj ptr:0x18 (BE)]:

<>
rx_adj_field: 0x5300 Load balancing
UTurn_egress_ifhandle: 0x12000140
<> section (if any)
RX H/W Result on NP:1 [Adj ptr:0x18 (BE)]: Section repeats for
<>
every NP
Switch Fabric Port and Platform Interface Info
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh control pm int G0/7/0/2 loc 0/7/CPU0
Ifname(1): GigabitEthernet0_7_0_2, ifh: 0x12000140 :

iftype 0xf
Matches egress_ifhandle
egress_uidb_index 0x5 from previous slide
ingress_uidb_index 0x5
port_num 0x2
subslot_num 0x0
phy_port_num 0x2 Matches rx_adj_field from previous slide
channel_id 0x3
channel_map 0x0 Ports connected to the same NPU share the
lag_id 0x0
same SFP value
virtual_port_id 0x0
switch_fabric_port 0x53
<>
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#show uidb index location 0/7/CPU0 gig0/7/0/2
-------------------------------------------------------------------------------------------------------
Location Interface-handle Interface-name Interface-Type Ingress-index Egress-index
-------------------------------------------------------------------------------------------------------
0/7/CPU0 0x12000140 GigabitEthernet0_7_0_2 Main interface 5 5
Show commands
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh cef 10.1.2.0/24 hardware egress location 0/7/CPU0
10.1.2.0/24, version 256779, internal 0x4004001 (ptr 0xa4d3ee6c) [1], 0x0 (0xa35bdc80), 0x440 (0xa5b1bd50)
Updated Jan 29 05:54:26.189
local adjacency 40.0.3.1
path-idx 0 [0xa59683c8 0xa5a30268]
next hop 40.0.3.1
local adjacency
local label 17012 labels imposed {ImplNull}
LEAF - HAL pd context :
sub-type : IPV4, ecd_marked:0, has_collapsed_ldi:0, collapse_bwalk_required:0, ecdv2_marked:0
Leaf H/W Result:
<>
fast_switch_entry
rx_adj_field: 0x5300 (LE) LE == Little Endian
egress_ifhandle: 0x12000140 (LE) 0xdc05 (LE) 0x05dc = 1500
<>
TX H/W Result for NP:0 (index: 0x82c (BE)):
<>
uidb_index : 0x500 (LE)
l3_mtu : 0xdc05 (LE)
prefix_adj_cnt_index: 0x0
dest_mac : 0x001b.53ff.9a99 u1.reserved : 0123456789ab
<>
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh arp gig0/7/0/2 loc 0/7/CPU0
Address Age Hardware Addr State Type Interface

40.0.3.1 00:43:01 001b.53ff.9a99 Dynamic ARPA GigabitEthernet0/7/0/2
Show commands (GRE tunnel example)
RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh cef 220.0.0.52/32 hardware egress location 0/7/CPU0

220.0.0.52/32, version 256780, internal 0x4004001 (ptr 0xa4d1f4bc) [1], 0x0 (0xa644e520), 0x440 (0xa5b1bd80)
Updated Jan 29 06:30:54.170
local adjacency point2point
via 41.52.0.2, tunnel-ip52, 5 dependencies, weight 0, class 0 [flags 0x0]
path-idx 0 [0xa59cc25c 0x0]
next hop 41.52.0.2
local adjacency
local label 16005 labels imposed {None} interface tunnel-ip52
<>
TX H/W Result for NP:0 (index: 0x1421 (BE)): ipv4 address 41.52.0.1 255.255.255.252
<> tunnel source Loopback0
uidb_index : 0x6100 (LE) tunnel destination 110.0.0.52
l3_mtu : 0xc405 (LE)
prefix_adj_cnt_index: 0x27670300
GRE Adj:
ip_sa : 110.0.0.41 ip_da : 110.0.0.52
tos : 0 ttl : 0xff
df : 1 tos_reflect : 1
rsvd flag bits : 0
encap_checksum : 0xe471
L3 Unicast Forwarding
Packet Flow (Simplified)
from wire
Rx LAG hashing
lookup key LAGID SFP Packet rewrite
LAG
L3: (VRF-ID, IP DA) System headers added
ECH Type:
TCAM rxIDB L3FIB rx-adj rewrite L3_UNICAS
SFP T
Packet Source L3 FIB Next-hop Switch Fabric Port
classification interface info lookup (egress NPU)
ACL and QoS Lookup SFP
Ingress NPU also happen in parallel
Fabric
Tx LAG hashing
LAG
ECH Type:
txIDB tx-adj L3FIB L3_UNICAST
rewrite
destination Next-hop L3 FIB => L3FIB lookup
interface info lookup
ACL and QoS Lookup

Egress NPU happens before rewrite
to wire ECH type: tell egress NPU type of lookup it should execute
CEF HW Resources Monitoring
RP/0/RSP0/CPU0:A9K-1#sh cef resource hardware ingress detail location 0/7/CPU0
CEF resource availability summary state: GREEN
CEF will work normally
ipv4 shared memory resource:
CurrMode GREEN, CurrAvail 247840769 bytes, MaxAvail 682823680 bytes
ipv6 shared memory resource:
mpls shared memory resource:
common shared memory resource:
DATA_TYPE_TABLE_SET hardware resource: GREEN
DATA_TYPE_TABLE hardware resource: GREEN
DATA_TYPE_IDB hardware resource: GREEN
DATA_TYPE_IDB_EXT hardware resource: GREEN
<>
ECMP Load balancing IPv6 uses first 64 bits in 4.0
releases, full 128 in 42
A: IPv4 Unicast or IPv4 to MPLS (3) releases
No or unknown Layer 4 protocol: IP SA, DA and Router ID (sh arm router-ids)

UDP or TCP: IP SA, DA, Src Port, Dst Port and Router ID
B: IPv4 Multicast
For (S,G): Source IP, Group IP, next-hop of RPF
For (*,G): RP address, Group IP address, next-hop of RPF
C: MPLS to MPLS or MPLS to IPv4
# of labels <= 4 : if payload is IP: same as IPv4 unicast; EoMPLS: etherheader will follow: innermost label + RID)
# of labels > 4 : 4th label and Router ID
- L3 bundle uses 5 tuple as A (eg IP enabled routed bundle interface)

- MPLS enabled bundle follows C
- L2 access bundle uses access S/D-MAC + RID, OR L3 if configured (under l2vpn)
- L2 access AC to PW over mpls enabled core facing bundle uses PW label (not FAT-PW label even if configured)
- FAT PW label only useful for P/core routers
Load-balancing scenarios
EoMPLS protocol stack
MPLS/IP protocol stack
45 for ipv4
MPLS vs IP Based loadbalancing
When a labeled packet arrives on the interface.
The ASR9000 advances a pointer for at max 4 labels.
If the number of labels <=4 and the next nibble seen right after that label is
4: default to IPv4 based balancing
6: default to IPv6 based balancing
This means that if you have a P router that has no knowledge about the MPLS service of
the packet, that nibble can either mean the IP version (in MPLS/IP) or it can be the
DMAC (in EoMPLS).
RULE: If you have EoMPLS services AND macs are starting with a 4 or 6. You HAVE to
use Control-Word 45 (ipv4)
L2 MPLS MPLS 0000 (CW) 4111.0000.
41-22-33 (mac)
Control Word inserts additional zeros after the inner label showing the P nodes to go for
label based balancing.
In EoMPLS, the inner label is VC label. So LB per VC then. More granular spread for EoMPLS can be
achieved with FAT PW (label based on FLOW inserted by the PE device who owns the service
Loadbalancing ECMP vs UCMP and polarization
Support for Equal cost and Unequal cost
32 ways for IGP paths
32 ways (Typhoon) for BGP (recursive paths) 8-way Trident
64 members per LAG
Make sure you reduce recursiveness of routes as much as possible (static
route misconfigurations)
All loadbalancing uses the same hash computation but looks at different bits
from that hash.
Use the hash shift knob to prevent polarization.
Adj nodes compute the same hash, with little variety if the RID is close
This can result in north bound or south bound routing.
Hash shift makes the nodes look at complete different bits and provide more spread.
Trial and error (4 way shift trident, 32 way typhoon, values of >5 on trident result in modulo)
cefBRKSPG-2904
load-balancing algorithm adjust <n>
2013 Cisco and/or its affiliates. All rights reserved. Cisco Public 88
Equal ADJ depth is UCMP requirement ADJ
1G
IGP Hop 1
ADJ
BGP Next Hop 1
1/6th 172.16.1.2
1G IGP Hop 1
Prefix
198.18.7.2 ADJ
100.100.100.0/24
1G IGP Hop 2
5/6th BGP Next Hop 2
3.3.3.3 5G 198.18.7.6 ADJ
Default is ECMP 3/6 vs 3/6
But DMZ link BW can balance 1G IGP Hop 3
unequally between the 2 paths 198.18.7.10 ADJ
1G IGP Hop 4
ECMP between paths for BGP ADJ
next hop. These have to be
198.18.7.14
NON recursive 1G IGP Hop 5
198.18.7.18 ADJ
Show CEF output for loadbalancing
Unequal adj depth breaks loadbalancing
RP/0/RSP0/CPU0:PR-ASR9K-3#show cef 3.3.3.3/32 detail capabilities
Tue Apr 23 08:27:41.826 UTC
3.3.3.3/32, version 611, internal 0x4000001 (ptr 0x7178e220) [4], 0x0 (0x0), 0x0 (0x0)
Updated Apr 23 08:27:23.875
gateway array (0x70f2524c) reference count 1, flags 0x8020, source rib (5), 0 backups
[1 type 3 flags 0x90111 (0x7105025c) ext 0x0 (0x0)] Buckets for LB
LW-LDI[type=0, refc=0, ptr=0x0, sh-ldi=0x0] distribution and
router static path index
address-family ipv4 unicast
Level 1 - Load distribution: 0 1 2 3 4
3.3.3.3/32 198.18.7.2
[0] via 198.18.7.2, recursive 3.3.3.3/32 198.18.7.6
3.3.3.3/32 198.18.7.10
[1] via 198.18.7.6, recursive 3.3.3.3/32 198.18.7.14
3.3.3.3/32 198.18.7.18
[2] via 198.18.7.10, recursive
[3] via 198.18.7.14, recursive Static routes missing a next hop interface are
perceived recursive!!
Non recursive static routes
RP/0/RSP0/CPU0:PR-ASR9K-3#show cef 3.3.3.3/32 det
RP/0/RSP0/CPU0:PR-ASR9K-3#show cef 3.3.3.3 detail loc 0/0/cpu0
3.3.3.3/32, version 4471, internal 0x4000001 (ptr 0x8850f79c) [4], 0x0 (0x0), 0x 3.3.3.3/32, version 695, internal 0x4000001 (ptr 0x7178e220) [7], 0x0
..
..
via 198.18.7.2, GigabitEthernet0/0/0/5.10, 4 dependencies, weight 0, class 0
[flags 0x0]
Level 1 - Load distribution: 0 1 2 3 4
path-idx 0 [0x7213a560 0x0]
[0] via 198.18.7.2, recursive next hop 198.18.7.2
remote adjacency
router static via 198.18.7.6, GigabitEthernet0/0/0/5.20, 4 dependencies, weight 0, class 0
[2] via 198.18.7.10, recursive address-family ipv4 unicast path-idx 1 [0x7213a5bc 0x0]
[3] via 198.18.7.14, recursive 3.3.3.3/32 198.18.7.2 next hop 198.18.7.6
3.3.3.3/32 198.18.7.6 remote adjacency
[4] via 198.18.7.18, recursive 3.3.3.3/32 198.18.7.10 .
3.3.3.3/32 198.18.7.14
3.3.3.3/32 198.18.7.18
Load distribution: 0 1 2 3 4 (refcount 2)
Hash OK Interface Address

router static 0 Y GigabitEthernet0/0/0/5.10 remote
address-family ipv4 unicast 1 Y GigabitEthernet0/0/0/5.20 remote
3.3.3.3/32 GigabitEthernet0/0/0/5.10 198.18.7.2 2 Y GigabitEthernet0/0/0/5.30 remote
3.3.3.3/32 GigabitEthernet0/0/0/5.40 198.18.7.14
3.3.3.3/32 GigabitEthernet0/0/0/5.50 198.18.7.18
Show cef for recursive prefix (non fixed)
Weight distribution:
slot 0, weight 9, normalized_weight 5
slot 1, weight 9, normalized_weight 5 Weight is 5 (5 next hops for 1 prefix)
Level 1 - Load distribution: 0 1 0 1 0 1 0 1 0 1 10 indexes, because weight is 5 and 2 paths

via 3.3.3.3, 4 dependencies, recursive, bgp-ext, bgp-multipath [flags 0x60a0] via 172.16.1.2, 15 dependencies, recursive, bgp-ext, bgp-multipath [flags 0x60a0]
path-idx 1 [0x7178f078 0x0]
path-idx 0 [0x7178e220 0x0] next hop 172.16.1.2 via 172.16.1.2/32
next hop 3.3.3.3 via 3.3.3.3/32
Load distribution: _ _ _ _ _ _ _ _ _ _ (refcount 1)
Adj is remote because
Show command not done with location 0/0/CPU0
Hash OK Interface Address
- Y GigabitEthernet0/0/0/5.50 remote
- Y GigabitEthernet0/0/0/0 remote
- Y GigabitEthernet0/0/0/5.10 remote - Y GigabitEthernet0/0/0/0 remote
- Y GigabitEthernet0/0/0/0 remote
- Y GigabitEthernet0/0/0/0 remote 50/50 split over 2 paths
- Y GigabitEthernet0/0/0/5.30 remote - Y GigabitEthernet0/0/0/0 remote
Show cef for the recursive prefix (fixed)
Weight distribution:
slot 0, weight 9, normalized_weight 9 This weight is set as part of the dmz link BW
slot 1, weight 1, normalized_weight 1
(not auto computed!!)
Level 1 - Load distribution: 0 1 0 0 0 0 0 0 0 0
via 3.3.3.3, 7 dependencies, recursive, bgp-ext, bgp-multipath [flags 0x60a0] via 172.16.1.2, 7 dependencies, recursive, bgp-ext, bgp-multipath [flags 0x60a0]
path-idx 1 [0x7178f078 0x0]
path-idx 0 [0x7178e220 0x0] next hop 172.16.1.2 via 172.16.1.2/32
next hop 3.3.3.3 via 3.3.3.3/32
Load distribution: 0 (refcount 1)
Load distribution: 0 1 2 3 4 (refcount 1)
Hash OK Interface Address Hash OK Interface Address
5 Y GigabitEthernet0/0/0/0 remote
0 Y GigabitEthernet0/0/0/5.10 remote
Great references
Understanding NP counters
https://supportforums.cisco.com/docs/DOC-15552
Capturing packets in the ASR9000 forwarding path
Loadbalancing Architecture for the ASR9000
Understanding UCMP and ECMP
Troubleshooting ASR9000 Forwarding: Multicast
Software Architecture MRIB/MFIB
IGMP
MRIB RP
PIM
MFIB PI MFIB PI MFIB PI
MFIB PD MFIB PD MFIB PD
LC1
LC0 LC2
Software Architecture MFIB on LC
LC MFIB PI
MFIB PD
NETIO PRM/uIDB MGID Server
NP Fabric
MGIDs and FGIDs
MGID - Multicast Group Identifier
Unique ID assigned to a multicast group
Used by FIA/Bridge to determine replication requirements per multicast group
FGID - Fabric Group Identifier
Slotmask used by switch fabric to determine replication to line card/RSP slots
Assigned to each group by multicast PD control plane
FGID (Slotmask)
FGIDs: 10 Slot Chassis FGIDs: 6 Slot Chassis
Phy
Logical
Slot
Slot
Number Logical
5 Slot
LC 3
4 LC 2
LC 7
LC 6
LC 5
LC 4
RSP 0
RSP 1
LC 3
LC 2
LC 1
LC 0
3 LC 1
2 LC 0
9 8 7 6 5 4 3 2 1 0 1 RSP 1
0 RSP 0
Slot Slot Mask
Logical Physical Binary Hex
LC3 5 0000100000 0x0020
LC2 4 0000010000 0x0010

Slot Slot Mask
LC1 3 0000001000 0x0008
Logical Physical Binary Hex
LC0 2 0000000100 0x0004
LC7 9 1000000000 0x0200
RSP1 1 0000000010 0x0002
LC6 8 0100000000 0x0100
LC5 7 0010000000 0x0080 RSP0FGID Calculation
0 Examples
0000000001 0x0001
LC4 6 0001000000 0x0040

RSP0 5 0000100000 0x0020
Target Linecards FGID Value (10 Slot Chassis)
RSP1 4 0000010000 0x0010 LC6 0x0100

LC3 3 0000001000 0x0008
LC2 2 0000000100 0x0004 LC1 + LC5 0x0002 | 0x0080 = 0x0082
LC1 1 0000000010 0x0002

LC0 + LC3 + LC7 0x0001 | 0x0008 | 0x0200 = 0x0209
LC0 0 Cisco Systems,0000000001
2006 Inc. All rights reserved. Cisco0x0001
Confidential EDCS:xxxx 5
MGID Tables MGID Bitmasks
MGID Bit 1 Bit 0

FIA
MGID Bit 1 Bit 0

Bridge1 Bridge0 MGID Bit 1 Bit 0
NP3 NP2 NP1 NP0
MGID Tables Mcast traffic replication based on mgid
MGID 1 0
FIA
MGID 1 0
Bridge1 Bridge0 MGID 0 0
NP3 NP2 NP1 NP0
show controller fabric
show controller fabric
fia stats
fia bridge stats
Trident LC Typhoon LC
NP3
3x10G
NP1 E
SFP +
RSP
FIA 3x 3x10G
Br1 NP2 10G E
SFP +
NP2 3x 3x10G
NP3 10G E
SFP +
FIA 3x 3x10G
NP4 10G E
Switch Fabric
FIA 3x
SFP +
FABRIC 10G
3x10G
NP1 NP5 E
ASIC
FIA 3x
SFP +
3x10G
Br0 NP6 10G E
SFP +
3x 3x10G
NP7 10G E
NP0
controller np ports all loc <>
FIA 3x
SFP +
3x10G
controller np ports all loc <> NP8 10G E
Show controller np counters <np> loc <Ingress NP:
Show controller np counters <np> loc < SFP +
ENET RX From Port
Egress NP:
FAB TX To fabric ENET FAB RX From Fabric
IPv4MC_DO_ALL_BUT_FRD Punt only FAB TX to TM
IPv4MC_DO_ALL punt to LC CPU LOOPBACK RX from TM
IFIB IGMP, PIM Control packets ENET TX to port
L3 Multicast Show CLIs
L2 Multicast Show CLIs
sh igmp snoop sum
show l2vpn forward mroute ipv4 hardware sh igmp snoop sum stat
sh igmp snoop group
sh igmp snoop bridge
LC1
T3
IGMP B0 Switch Fabric
Interface
T2
Fabric
1
T1 IGMP Snooping
B1
T0 2
3
4 L2FIB L2FIB RP
(142.0.0.2, (142.0.0.2, (142.0.0.2,
225.0.0.1/225.0.0.2) 225.0.0.1/225.0.0.2) 225.0.0.1/225.0.0.2)
Receiver
Gig0/4/0/2
Gig0/4/0/10.101 Source
Join 225.0.0.1, Ten0/5/0/1
Gig0/4/0/10.100
225.0.0.2
Receiver
Gig0/4/0/3.102 igmp v2
Join 225.0.0.1
multicast-routing
interface GigabitEthernet0/4/0/10.101 address-family ipv4
ipv4 address 33.0.2.1 255.255.255.0 interface all enable
encapsulation dot1q 101 router pim
interface GigabitEthernet0/4/0/3.102 address-family ipv4
ipv4 address 42.0.1.2 255.255.255.0
encapsulation dot1q 102
rp-address 110.0.0.24
interface TenGigE0/5/0/1 interface GigabitEthernet0/4/0/2
ipv4 address 40.0.75.2 255.255.255.0 enable
! interface GigabitEthernet0/4/0/10.100
multicast-routing
enable
address-family ipv4
interface all enable RP/0/RSP0/CPU0:ASR9K-2#
router pim
address-family ipv4
rp-address 110.0.0.24
interface TenGigE0/5/0/1
enable
interface GigabitEthernet0/4/0/3.102
enable
enable
RP/0/RSP0/CPU0:ASR9K-3#
Example 1 L3 Multicast PIM SSM
Show CLI Validate the mrib and mfib entry
RP/0/RSP1/CPU0:asr9k-2#show mrib route 225.0.0.1

== snip ==
(142.0.0.2,225.0.0.1) RPF nbr: 142.0.0.2 Flags: L
Up: 4d05h
Incoming Interface List
GigabitEthernet0/4/0/10.100 Flags: A, Up: 4d03h
Outgoing Interface List
GigabitEthernet0/4/0/2 Flags: F NS, Up: 2d22h
RP/0/RSP0/CPU0:asr9k-3#show mrib route 225.0.0.2 detail
=== snip ===
(142.0.0.2,225.0.0.2) Ver: 0x2fba RPF nbr: 40.0.75.1 Flags:,
PD: Slotmask: 0x40 Same slot mask as 225.0.0.1. Because egress LC is same.
MGID: 19921 MGID should be different (ackets replicated to only one NP).
Up: 2d23h
TenGigE0/5/0/1 Flags: A, Up: 2d23h
GigabitEthernet0/4/0/10.101 Flags: F NS, Up: 2d23h
RP/0/RSP0/CPU0:asr9k-3#
MGID tables Getting MGID and Displaying MGID table
RP/0/RSP0/CPU0:asr9k-3#show mrib route 225.0.0.1 detail

(*,225.0.0.1) Ver: 0x429a RPF nbr: 40.0.75.1 Flags: C,
PD: Slotmask: 0x40
MGID: 19919
Up: 2d21h
TenGigE0/5/0/1 Flags: A NS, Up: 2d21h
GigabitEthernet0/4/0/3.102 Flags: F NS LI, Up: 14:20:00
GigabitEthernet0/4/0/10.101 Flags: F NS LI, Up: 2d21h
(142.0.0.2,225.0.0.1) Ver: 0x7163 RPF nbr: 40.0.75.1 Flags:,
PD: Slotmask: 0x40 FGID Used for Fabric Replication 0x40 == 0001000000 (slot 4)
MGID: 19918 MGID Used by egress LCs FIA and Bridge ASIC for replication
Up: 3d00h
TenGigE0/5/0/1 Flags: A, Up: 3d00h Interface towards source (RPF to source)
GigabitEthernet0/4/0/3.102 Flags: F NS, Up: 14:20:00 interface towards receivers
GigabitEthernet0/4/0/10.101 Flags: F NS, Up: 2d21h interface towards receivers
RP/0/RSP0/CPU0:asr9k-3#show controllers mgidprgm mgidindex 19918 location 0/4/CPU0
Device MGID-Bits Client-Last-Modified

=======================================================
FIA 10 MFIBV4 Replicated to Bridge-1 [Bridge-1 | Bridge-0]

Bridge-0 0 MFIBV4 Not replicated here [NP 1 | NP 0]
Bridge-1 11 MFIBV4 Replicated to NP 2 and 3 [NP 3|NP 2]
MGID/FGID and NP
RP/0/RSP0/CPU0:asr9k-3#show mfib hardware route olist 225.0.0.1 location 0/4/CPU0
------ SNIP----
Source: 142.0.0.2 Group: 225.0.0.1 Mask: 64 RPF Int: Te0/5/0/1
Route Information
------------------------------------------------------------------------
B S DC PL PR PF DR RI FS G M
------------------------------------------------------------------------
F F F F F F F 0xe000100 0x40 19918 3797 FGID and MGID values
------------------------------------------------------------------------
Interface Information
------------------------------------------------------------------------
NP Intf OT U T IC B
------------------------------------------------------------------------
2 Gi0/4/0/10.101 REG 85 1 F F NP and Outgoing port info
3 Gi0/4/0/3.102 REG 109 1 F F NP and Outgoing port info
------------------------------------------------------------------------
OLIST counts
--------------------------------------------------------
NP: 0 1 2 3
Count: 0 0 1 1 Shows 1 port from NP 2 and 3 interested in traffic.
--------------------------------------------------------
Legend to previous output
--------------------------------------------------------------------------
Legend:
Route Information
NP: NP ID B: BACL check
S: RPF Interface signal DC: Directly connected
PL: Punt to LC CPU PR: Punt to RP
PF: Punt if forwarded DR: Drop all
RI: RPF interface FS: Fabric slotmask
G: Multicast group ID M: Multicast Leaf Index
T: Table ID for lookup OC: Count of OLIST members
Base: Base of the statistics pointer NI: Not Installed
Interface Information
NP: NP ID Intf: Interface
U: uIDB index OT: OLE Type
T: Table ID IC: HW IC flag
B: HW BACL bit EU: Interface uIDB index
IB: Bundle interface EH: In HW OLIST table
OIDX: OLIST index on NP PT: Punt table entry
Base: Statistics Ptr base RM: Remote FGID (Pri/Back)
Software OLIST Information

SW OC: Software OLIST counts HW OC: Hardware OLIST counts
T: Table ID SD: Send direct flag
--------------------------------------------------------------------------
Example 1 L3 Multicast PIM SM
show CLI check the counters [1]
RP/0/RSP0/CPU0:asr9k-3#show mfib hardware route statistics 225.0.0.1 142.0.0.2 loc 0/5/CPU0
N: NP ID
LC Type: Typhoon A9K-MOD160-SE
Source: 142.0.0.2 Group: 225.0.0.1 Mask:64 R: Received
------------------------------------------------------------------------- F: Forwarded
NP R(packets:bytes)/F(packets:bytes)/P(packets)/ID(packets)/ED(packets) P: Punted to CPU
------------------------------------------------------------------------- ID: Ingress Drop
0 406759:18710914 / 0:0 / 0 / 0 / 0 THIS NP is receiving traffic from wire ED: Egress Drop
1 0:0 / 0:0 / 0 / 0 / 0
2 0:0 / 0:0 / 0 / 0 / 0
3 0:0 / 0:0 / 0 / 0 / 0
-------------------------------------------------------------------------
RP/0/RSP0/CPU0:asr9k-3#show mfib hardware route statistics 225.0.0.1 142.0.0.2 loc 0/4/CPU0
LC Type: Trident A9K-40GE-E
--------------------------------------------------------------------------
Source: 142.0.0.2 Group: 225.0.0.1 Mask:64
-------------------------------------------------------------------------
NP R(packets:bytes)/F(packets:bytes)/P(packets)/ID(packets)/ED(packets)
-------------------------------------------------------------------------
0 0:0 / 0:0 / 0 / 0 / 0
1 0:0 / 0:0 / 0 / 0 / 0
2 0:0 / 434208:19973568 / 0 / 0 / 0 This NP is sending traffic out on wire
3 0:0 / 443309:20392214 / 0 / 0 / 0 This NP is sending traffic out on wire
-------------------------------------------------------------------------
Interface Statistics:
-------------------------------------------------------------------------
C Interface F/P/D (packets:bytes)
-------------------------------------------------------------------------
2 Gi0/4/0/10.101 434208:19973568 / 0:0 / 0:0 Outgoing interface on the NP2
3 Gi0/4/0/3.102 443309:20392214 / 0:0 / 0:0 Outgoing interface on the NP3
-------------------------------------------------------------------------
QOS architecture
111
System QoS Refresh
End-to-End priority propagation Guarantee bandwidth, low latency for highOne
priority
Queuetraffic
set per each
at any congestion point NP on the LC
Ingress side of LC Egress side of LC
CPU CPU 4
PHY NP0 NP0 PHY
3
PHY NP1 2 NP1 PHY
FIA FIA
PHY NP2 Switch NP2 PHY
Fabric
PHY NP31 NP3 PHY
2 3
1 Virtual Egress FIA 4
Ingress (sub-)interface Output Queues Egress (sub-)interface
QoS Queues Queues QoS Queues
Implicit Configuration
Configure with Two strict high priority + Configure with Egress MQC
Ingress MQC 4-layer hierarchy Normal priority 4-layer hierarchy
Two strict high priority + Normal priority Two strict high priority + Normal priority
System QoS Refresh
Fabric Bandwidth Access Overview
3 strict priority scheduling/queueing
Back pressure and virtual output queue
Multicast and Unicast separation (separated
RSP0
Crossbar queues and fabric plane)
Fabric 5: credit return
ASIC
1: Fabric Request
Crossbar
Fabric Egress LC
Ingress LC
ASIC
Arbiter
FIA FIA
2: Arbitration
Crossbar
Fabric
3: Fabric Grant ASIC
Crossbar
4: load-balanced Fabric
ASIC
transmission across
fabric links Arbiter
RSP1
Arbitration & Fabric QoS
Arbitration is being performed by a central high speed arbitration ASIC on the
RSP
At any time a single arbiter is responsible for arbitration (active/active APS
like protection)
The Arbitration algorithm is QOS aware and will ensure that P1 classes have
preference over P2 classes, both of which have preference over non-priority
classes
Arbitration is performed relative to a given the egress VQI
System QoS Refresh (3)
Backpressure and VoQ Mechanism
Egress NP congestion backpressure to ingress FIA

Packet is en-queued in the dedicated VoQ
No impact of the packet going to different egress NP
One VoQ set (4 queues)
per each NP in the system No head-of-line-block issue
Backpressure: egress NP egress FIA
fabric Arbiter ingress FIA VoQ
Ingress side of LC1 Egress side of LC2
10Gbps CPU CPU 1

PHY NP0 NP0 PHY 2
5Gbps
PHY
5Gbps
NP1 NP1 PHY
FIA FIA
PHY NP2 Switch NP2 PHY 3
Fabric
PHY NP3 NP3 PHY
Packet going to different egress NP put into different VoQ set Congestion on
BRKSPG-2904 one NP wont block
2013 Cisco the
and/or its packet
affiliates. going
All rights reserved.to different NP Cisco Public 115
Linecard QoS
Switch Fabric Queuing mechanisms
Slot 0/Port 0
DRR
DRR
Port 0
Slot 0/Port 1
DRR
DRR
Ingress Fabric Scheduler
Egress Fabric Scheduler

Port 1
. RSP 0
.
DRR
Port 2
DRR
Slot 9/Port 2
DRR
DRR
Port 3
Slot 9/Port 3
RSP 1
DRR
DRR
Multicast Egress
Switch Fabric Fabric ASIC
Multicast
136 ingress VoQ used: 20 egress fabric queues:

8 dest LCs * 4 10G ports/LC * 4 classes/port** == 128 VoQ for LCs 4 classes/port * 4 ports/LC (unicast) == 16
2 dest RSPs * 1 10G port/RSP * 4 classes/port == 8 VoQ for RSPs 4 multicast classes == 4
4 multicast queues
BRKSPG-2904 higher density cards will
2013 Cisco have
and/or correspondingly
its affiliates. All rights reserved. larger numbers
Cisco Public of VoQs 116
Default Interface Queues
For every physical port in the system, the following queues get created:
Typhoon Trident
Ingress (NP->Fabric) Egress (NP->Line) Ingress (NP->Fabric) Egress (NP->Line)
High priority 1 queue High priority 1 queue (routing High priority queue (routing High priority queue (routing
(routing control protocols) control protocols and critical control protocols) control protocols and critical
traffic like BFD) traffic like BFD)
Medium priority 2 queue Medium priority 2 queue Medium priority queue Medium priority queue (Unused
(Unused without policy-map) (Unused without policy-map) (Unused without policy-map) without policy-map)
Medium priority 3 queue Medium priority 3 queue
(Unused without policy-map) (Unused without policy-map)
Low priority queue (Used by Low priority queue (Used by all Low priority queue (Used by Low priority queue (Used by all
all other traffic) other traffic) all other traffic) other traffic)
show qoshal default-queue subslot <n> port <m> location <location>

MQC to System QOS mapping
ASR 9000 supports traffic differentiation at all relevant points within the system
P1/P2/LP differentiation or P1/LP differentiation support throughout the system
Classification into these priorities is based on input MQC classification on the ingress linecard
into P1, P2, Other
Once a packet is classified into a P1 class on ingress it will get mapped to PQ1 queue along
the system qos path
Once a packet is classified into a P2 class on ingress it will get mapped to PQ2 queue along
the system qos path, unless no MP is implemented. In this case HP would be used for P2.
Once a packet is classified into a non-PQ1/2 class on ingress it will get mapped to LP queue
along the system qos path
Note: The marking is implicit once you assign a packet into a given queue on ingress; its sets
the fabric header priority bits onto the packet.
no specific set action is required
Feature order on ASR 9000 NP
(simplified)
TCAM
Ingress linecard
From wire I/F ACL QOS
Fwd lookup
classification classification classification
IFIB action QoS action L2 rewrite ACL action IFIB lookup
To fabric From fabric

egress linecard
QOS ACL
ACL action classification classification L2 rewrite Fwd lookup
QoS action To wire

Feature order on ASR 9000 NP
QoS Action Order
Ingress linecard
From wire I/F ACL QOS
Fwd lookup
classification classification classification
IFIB action QoS action L2 rewrite ACL action IFIB lookup
To fabric QoS Action From fabric

WRED classifies on marked/remarked values
egress linecard
QOS
(doesntACLswitch class-maps!)
ACL action classification classification L2Queue/sha
rewrite Fwd lookup
Police Mark pe/WRED
QoS action To wire
Injected packets
In general are injected to-wire (same as Pak Priority in IOS)
Means that all features are bypassed.
Including QOS
Few exceptions
ICMP
BFD echo responses
Netflow
ASR 9000 QOS Implicit Trust
For Bridged packets on ingress outermost COS would be treated

as trusted.
For Routed packets on ingress DSCP/Precedence/outermost
EXP would be treated as trusted based on packet type.
Default QOS will be gleaned from ingress interface before QOS
marking is applied on the ingress policymap.
By default ASR 9000 would never modify DSCP/IP precedence of
a packet without a policy-map configured.
Default QOS information would be used for impositioned fields only
ASR 9000
Linecard/NP QoS
Overview
Typhoon System QoS Overview
Typhoon system (new fabric, new LC) has the same internal system
qos and back pressure mechanism as existing system.
On Trident LCs, VoQ and FIA egress queue set is per NP basis.
NP is 1:1 for 10GE ports
On the new LC system, NP is designed for multiple 10G ports, 40G,
and 100G port. sets of VQIs are used to represent 10/40/100G ports
Each 10G port is 1:1 mapped to one VQI
Each 40G port is mapped to 8 VQI
Each 100G port is mapped to 16 VQI
VQIs used to load balance across internal connections
Typhoon QoS Overview
Super-set of existing Trident linecard QoS functionality
Dedicated TM for queuing
Fabric/internal QoS mechanism
Flexible 4-level H-qos ingress and egress
Higher scale
Higher queue and policer scale
More granular bandwidth control for both policing and queuing
Higher buffer size
Additional new feature capability
Conform-aware policer (a/k/a Coupled Policer)
4 strict priority: P1, P2, P3 (egress only) and normal priority
P3 is egress only; need special consideration in case of parent shaper
Ingress TM for <=30G configs only

No input shaping on high-NP loading configs (36x10G, 8x10 MPA, 40G MPA)
By default all queue-limits are set to 100ms of service rate
Increased Priority Queues
Trident Max of 8 Child Queues per parent , with 1 Priority 1, 1 Priority 2, and 6
Normal-priority queues (including class-default)
Typhoon Max 8 Child Queues per Parent Choices based on user config in
policy.
1 Priority 1, 2 Priority 2 and 5 Normal-priority
1 Priority 1, 1 Priority 2, 1 Priority 3, 5 Normal-Priority (Egress only)
1 Priority 1, 1 Priority 2, and 6 Normal-priority
ASR 9000 QOS
Functional Details
ASR9K QoS Classification Criteria
Very flexible L2/L3 field classification on L2 interfaces
Inner/outer cos
Inner/Outer vlan *
DEI*
Outer EXP
Dscp/Tos
TTL, TCP flags, source/destination L4 ports
Protocol
Source/Destination IPv4
Source/Destination MAC address*
Discard-class
Qos-group
match all/match any
Note:
Not all fields are supported on L3 interfaces*
Some fields dont make sense on ingress (e.g. dicard-class, qos-group)
MPLS classification is based on EXP only
ASR9K QoS - Classification Formats
Per Policy-map a given classification format is chosen by SW, i.e a given
policy-map can only classify based on a single format
Format 0 Format 1 Format 2 Format 3
Fields supported -IPV4 source address -Outer VLAN/COS/DEI -Outer VLAN/COS/DEI -Outer VLAN/COS/DEI
(Specific/Range)[1] -Inner VLAN/COS -Inner VLAN/COS -Inner VLAN/COS
-IPV4 Destination address -IPV4 Source address -IPV4 Destination address -MAC Destination address
(Specific/Range) (Specific/Range) (Specific/Range) -MAC source address
-IPV4 protocol -IP DSCP / TOS / Precedence -IP DSCP / TOS / Precedence -QOS-group (output policy
-IP DSCP / TOS / Precedence -QOS-group (output policy only) -QOS-group (output policy only) only)
-IPV4 TTL -Discard-class (output policy only) -Discard-class (output policy only) -Discard-class (output
-IPV4 Source port (Specific/Range) policy only)
-IPV4 Destination port
(Specific/Range)
-TCP Flags
-QOS-group (output policy only)
-Discard-class (output-policy only)
ASR9K QoS - Packet marking details
settable packet fields:
dscp/precedence
EXP imposition
EXP topmost
cos inner/outer
qos-group
discard-class
ASR9K supports maximum of 2 fields per class-map. The

same 2 fields can be placed in any combination below
- 2 sets per police-conform/exceed/violate
- 2 sets without policing.
Note: In MPLS context only EXP marking is supported
Remember that mpls encapped packets cant match on L3 criteria (same for ACL)
ASR9K QoS - Policing details
RFC 2698 supported (2r3c) and ingress/egress) on extended
1r2c linecards
Ingress & egress policing supported Policer actions supported:
General Rule: Policing required on transmit
priority queues. drop
Priority level 2 classes can also accept set (implicitly behaves like set and
shaping instead of policing. transmit)
Granularity of 8Kbps supported each color can have two set actions:
(typhoon, 64k on trident)
Policy-map parent
2-level nested policy maps Class class-default
supported Police rate 10 Mbps peak-rate 20 mbps
conform-action set dscp af12
Note: policers at parent and child work conform-action set cos 2
independently exceed-action set dscp af13
64k policers per NP (shared for exceed-action set cos 3
Normal Hierarchical Policer
policy-map child
class class1
police rate 20 mbps peak-rate 50 mbps
class class2
At parent level, if its over the CIR,
policy-map parent packet will be dropped randomly.
class class-default There is no awareness which packet
police rate 60 mbps to be dropped
service-policy child
Conform Aware Policer
policy-map child
class class1
class class2
Parent CIR must > aggregated child CIR

policy-map parent Parent police only support 1R2C, child police
class class-default support all: 1R2C, 2R3C, or 1R3C
service-policy child
police rate 60 mbps If drops happen at parent level, it will drop
child-conform-aware child out-of-profile packet, but guarantee the
child in-profile packet
Common Policer problems
Note that all L2 headers are included, added to the payload and that packet
size is depleting the token bucket (applies to shaping also). Only IFG and CRC
are not accounted for.
Incorrect burst size configuration, allow for some excess burst to catch up.
Mistake between 2 or 3 rate policers (exceed action drop)
Tridents policer cant go negative, Typhoon can borrow
This means that policer behavior is slightly different between the 2 hardware
ASR 9000 QoS - Queue scheduling
Use shape for a shaped PIR for a graceful enforcement of a maximum
bandwidth
shaping at all configurable levels
Min. granularity: 64kbps (L3, L4, 256kbps for L2)
priority levels: priority level 1/2/3, minBw/CIR and Bw remaining
Use bandwidth (minBw) for a CIR guarantee relative to the parent
hierarchy level
Min. RATE: 64kbps (8k granularity)
Use bandwidth remaining ratio/percent for the redistribution of excess
bandwidth that is available after PQ classes have been scheduled
configurable ratio values 1-1020
Two parameter scheduler support at class level and subscriber group level
(L4, L2):
Shape & BwR (ratio / percent)
Shape & MinBw (absolute / percent)
Not supported: BwR & MinBw on the same class
ASR 9000 QoS - congestion management/buffering details
WRED based on: DSCP, IPP, EXP, COS, discard-class

default queue-limit -to prevent buffer exhaustion- is 100ms of service rate
(service rate is the sum of guaranteed bw/bwr assigned to a class)
WRED configuration unit options are: bytes, kbytes, mbytes, us, ms,
packets
These values will be rounded up to a set of pre-defined profiles ranging from 8 kB to
262144 kB
The actual implementation uses 512 byte buffer particles
Novelty: ASR 9000 supports WRED on shaped PQ2 classes.
Can be used for differentiation of two kinds of priority within the PQ2 class
Absolute vs Percentage
All relevant policy actions support both, absolute and percentage based
configuration:
shape
bandwidth
Police
bandwidth remaining*
For tri-rate Copper SFPs (10/100/1000) percentage based QOS will be
adjusted automatically based on the selected rate
*Note: Bandwidth remaining supports ratio/percent, not absolute bandwidth
Show/debug QOS commands
show running-config
show running-config policy-map <policyname> Policy map configuration
show running-config class-map <classmap> Class map configuration
show running-config interface <interface> Interface running configuration
show policy-map interface <interface> [iNPt | output] Policy-map statistics on a particular non-bundle interface
show policy-map interface <bundle-interface> [iNPt|output]

Policy-map statistics on a member of bundle interface
member
show qos interface <interface> <iNPt|output> [member Displays hardware and software configured values of each class for
<interface>] a service-policy on an interface
Displays the detailed information of hardware and software
show qos-ea interface <interface> <iNPt|ouput> [member
configured paramters in each class of a service-policy on an
<interface>] [detail]
interface
show qos summary <police|policy|queue> [interface <interface>]
Lists the summary of all queues or policers or interfaces for a policy
[output|iNPt] [member <interface>]
show qoshal tm-config
<all|counters|fcu|general|priority|shape|topology|wfq|wred> np Displays generic NP TM config
<np> tm <tm>
show qoshal <wfq|wred|wred-scale|shape|police|police-node> np
Displays various profiles configured in sw and hw and the values of
<np> tm <tm> level <level> profile <profile> <num-of-profiles>
each profile
[hw|sw]
Show/debug QOS commands - contd
show qoshal default-queue subslot <n> port <m> location
Displays the default-queue information
<location>
Displays the summary of all the resources used in hardware and

show qoshal resource summary [np <np>]
software for QoS such number of policy instances, queues, profiles
show qoshal fcu <limits|status|profile> Displays all Traffic Manager (TM) Flow control related info
show qoshal ha chkpt <all|<chkpt-tbl-name> {all|<recid>|info} Display HA related info for PRM QoS HAL
Displays the HA State of process QoS EA whether it can accept the

show qos-ea ha state
service-policies
show qos-ea ha chkpt <all|<chkpt-tbl-name> {all|<recid>|info} Display HA Chkpt related info for all the chkpt tables for QoS EA
Displays the trace of errors or events or internal events of QoS EA

show qos-ea trace {all|errors|events|internal}
process
show prm server trace hal Displays all the trace info of PRM QoS HAL thread
debug qos-ea all Debug commands for qos ea process
debug qoshal <level|module|events> <word> Debug commands for PRM qos HAL
debug prm server hal <all|error|events> Debug commands for PRM qos HAL API
Troubleshooting Back-pressure Issues
Check if you are seeing FIA drops
RP/0/RSP1/CPU0:ios#show drops
Tue Jan 14 20:44:25.360 EST
Node: 0/0/CPU0:
<>
FIA 0 Drops:
----------------------------------------------------------------
Ingress Drops 287078960
Egress Drops 1
Total Drops 287078961
Ingress Generic Hard Drop-2 287078960
Egress Mcast RxFab Hdr-1 1
----------------------------------------------------------------
Check if any VQI is dropping packet

RP/0/RSP1/CPU0:ios#show controller fabric fia q-depth location 0/0/CPU0
FIA 0
VoQ | ddr | pri | pkt_cnt
------+-----+-----+---------
23 | 0 | 2 | 118
Total Pkt queue depth count = 118 Packets in the queue. Not good.
Troubleshooting Back-pressure Issues
Check if you are seeing FIA drops
RP/0/RSP1/CPU0:ios#show controllers pm loc 0/5/CPU0 | I ^Ifname|switch_fabric_port
Ifname(1): TenGigE0_5_0_0, ifh: 0xe000100 :
switch_fabric_port 0x17 VQI 23 is for interface ten0/5/0/0
Check egress NP TM Drops:

RP/0/RSP1/CPU0:ios#show controllers NP tm counters all location 0/5/CPU0
Node: 0/5/CPU0:
==== TM Counters (NP 3 TM 1) ====
TM Counters: commit_xmt_paks: 1509333316
excess_xmt_paks: 67641555690
Total Transmitted paks: 69150889006
wred_drop paks: 2441836834 timeout_drop 0 intf_drop 0
==== TM Counters (NP 3 TM 2) ====
TM Counters: commit_xmt_paks: 0
excess_xmt_paks: 0
Total Transmitted paks: 0
wred_drop paks: 0 timeout_drop 0 intf_drop 0
RP/0/RSP1/CPU0:ios#
What is programmed in HW?
RP/0/RSP0/CPU0:WEST-PE-ASR9K-2#sh qos interface g0/0/0/0 output
Fri Jan 10 15:05:40.347 EST
Interface: GigabitEthernet0_0_0_0 output policy-map p-map
Bandwidth configured: 500000 kbps Bandwidth programed: 500000 kbps
class class-default
ANCP user configured: 0 kbps ANCP programed in HW: 0 kbps
service-policy dummy
Port Shaper programed in HW: 500000 kbps
Policy: p-map Total number of classes: 2
shape average 500 mbps
---------------------------------------------------------------------- !
Level: 0 Policy: p-map Class: class-default end-policy-map
QueueID: N/A !
Shape Profile: 1 CIR: 64 kbps CBS: 10240 bytes PIR: 499968 kbps PBS: 6291456 bytes
WFQ Profile: 4 Committed Weight: 1 Excess Weight: 1
Bandwidth: 0 kbps, BW sum for Level 0: 0 kbps, Excess Ratio: 1 policy-map dummy
---------------------------------------------------------------------- class class-default
Level: 1 Policy: dummy Class: class-default bandwidth percent 100
Parent Policy: p-map Class: class-default !
QueueID: 642 (Priority Normal) end-policy-map
Queue Limit: 8388 kbytes Profile: 2 Scale Profile: 2
!
Bandwidth: 500000 kbps, BW sum for Level 1: 500000 kbps, Excess Ratio: 1
----------------------------------------------------------------------
What is programmed in HW?
RP/0/RSP0/CPU0:WEST-PE-ASR9K-2#sh
Rate is rounded to the nearest 8k orqos 64kinterface
value g0/0/0/0 output
Fri Jan 10 15:05:40.347 EST
Interface:
Shape sets PIR
GigabitEthernet0_0_0_0 output policy-map p-map
Bandwidth
PBS is default rate of500000
configured: 100msec
kbps of configured
Bandwidth shape rate
programed: 500000 kbps
class class-default
ANCP
ParentuserBW is zero or
configured: 64k,ANCP
0 kbps onlyprogramed
applicable in oversubscription
in HW: 0 kbps at sum of parent levels dummy
service-policy
Port Shaper programed in HW: 500000 kbps
Policy: p-map Total number of classes: 2
shape average 500 mbps
---------------------------------------------------------------------- !
Level: 0 Policy: p-map Class: class-default end-policy-map
QueueID: N/A !
Shape Profile: 1 CIR: 64 kbps CBS: 10240 bytes PIR: 499968 kbps PBS: 6291456 bytes
Bandwidth: 0 kbps, BW sum for Level 0: 0 kbps, Excess Ratio: 1 policy-map dummy
---------------------------------------------------------------------- class class-default
Level: 1 Policy: dummy Class: class-default bandwidth percent 100
Parent Policy: p-map Class: class-default !
QueueID: 642 (Priority Normal) end-policy-map
Queue Limit: 8388 kbytes Profile: 2 Scale Profile: 2
!
Bandwidth: 500000 kbps, BW sum for Level 1: 500000 kbps, Excess Ratio: 1
----------------------------------------------------------------------
Shaping with PIR/PBS and CIR
Shaper peaks to linerate for pbs time
Should allow some burst to get to PIR faster
CIR is ignored, will result in queue(exceed) counts, but they dont mean drops!
linerate
PBS
PIR
CIR
RP/0/RSP0/CPU0:A9K-BNG#show policy-map int g 0/0/0/0 | i Queue
Queueing statistics
Queue ID : 136
Queue(conform) : 0/0 0
Queue(exceed) : 0/0 0
QOS summary
All Ethernet linecards support Queuing, Marking and Policing.

Some high speed linecards do not support ingress Queuing (but
support policing and marking).
Because their ingress TM (Traffic Manager) is disabled
To guarantee priority end to end, make sure high priority traffic is
marked on ingress (This will not burn a queue)
Egress Line
ASR9K Default Marking card
IPP=5 DSCP=44
Ingress 802.1p = 1 *
Line card IPP=5 DSCP=44
802.1p = 1*
IPP=5 DSCP=44
L2 (VPWS,
802.1p = 1 VPLS,
internal
IPP=5 DSCP=44 802.1p= 1*
Bridging) cos = 1 EXP= 1*
IPP=5 DSCP=44
EXP= 1*
Carried in IPP=5 DSCP=44
internal
buffer
header IPP=5 DSCP=44
Note: VPWS will be treated like a L2 operation on ingress - Applies for all tags/labels in the stack that get imposed.
Not for VLAN translation. Bridging on egress without adding an vlan header is an hypothetical case in case we have a need.
IPP = IP Precedence, showing IPP & DSCP seperately since policymap can treat precedence and dscp separately as required.
ASR9K Default QOS
IPP=5 DSCP=44
L2 (VPWS,
Untagged VPLS,
internal 802.1p = 0 *
IPP=5 DSCP=44 IPP=5 DSCP=44
Bridging) cos = 0
802.1p = 0*
IPP=5 DSCP=44
802.1p= 0*
EXP= 0*
IPP=5 DSCP=44
EXP= 0*
IPP=5 DSCP=44
IPP=5 DSCP=44
Note: Trust cos in case of bridged interfaces in ingress. For untagged packets use cos = 0.
* - Applies for all tags/labels in the stack that get imposed.
ASR9K Default QOS
IPP=5 DSCP=44
Untagged
L2 (VPWS,
EXP=3 VPLS,
Bridging) cos = 0
802.1p = 0*
IPP=5 DSCP=44
802.1p= 0*
EXP= 0*
IPP=5 DSCP=44
EXP= 0*
IPP=5 DSCP=44
IPP=5 DSCP=44
Note: Trust cos in case of bridged interfaces in ingress. For untagged packets use cos = 0.
- Applies for all tags/labels in the stack that get imposed.
- Explicit NULL EXP is treated the same as
BRKSPG-2904 anCisco
2013 topmost EXP of
and/or its affiliates. nonreserved.
All rights NULL labels. Cisco Public 148
ASR9K Default QOS
IPP=5 DSCP=44
Routed
Interface
(IP/MPLS) cos = 5
802.1p = 5 *
IPP=5 DSCP=44
802.1p= 5 *
EXP= 5 *
IPP=5 DSCP=44
EXP= 5*
IPP=5 DSCP=44
IPP=5 DSCP=44
Note: Trust dscp in case of routed interfaces in ingress. For Non IP packets use cos = 0
ASR9K Default QOS
IPP=5 DSCP=44
Routed
802.1p = 1 Interface
(IP/MPLS) cos = 5
802.1p = 5 *
IPP=5 DSCP=44
802.1p= 5 *
EXP= 5 *
IPP=5 DSCP=44
EXP= 5*
IPP=5 DSCP=44
IPP=5 DSCP=44
Note: Trust dscp in case of routed interfaces in ingress. For Non IP packets use internal dscp= 0
ASR9K Default QOS
IPP=5 DSCP=44
802.1p = 1
Routed
EXP = 3 Interface
(IP/MPLS) cos = 3
802.1p = 3 *
IPP=5 DSCP=44
802.1p= 3 *
EXP= 3 *
IPP=5 DSCP=44
EXP= 3 *
IPP=5 DSCP=44
IPP=5 DSCP=44
Note: Trust EXP/dscp in case of routed interfaces in ingress. For Non IP packets use internal dscp= 0. Do not
overwrite DSCP fields exposed during disposition to support pipe mode by default.
* - Applies for all tags/labels in the stack that get imposed. 151
Scaled ACL
Scaled ACL problem statement:
Provide a solution that can do ACL filtering for exceptionally large rulesets at high
packet rates, within hardware (cost/power/space) constraints that makes it
affordable/ deployable, with low variability in performance.
Hot Tip: This is really #^(&ing hard. But we did it anyway.
Two part solution:
1. how do you configure really large rulesets in the
management plane
2. how do you process them in the data plane?
Configuration improvements:
Data structure selection:
Trees (tries): provide efficient memory usage, but non-deterministic (highly
variable) performance.
The number of lookups can vary a lot depending on exactly where you find
the match.
The Juniper MX solution builds the ACL rulesets into trees, which are then
stored in very fast (but very small) lookup memory, and used for forwarding.
TCAMs:
Essentially reverse memory that takes a lookup key and mask, and returns
a result. (TCAM rule or ValueMaskResult)
Always returns the result in a single memory access (i.e. order one
lookup) so its really fast and very determinstic.
BUT, TCAMs are large, dedicated hardware devices. High power, high
cost, and limited to (practically) tens of thousands of rules.
Configuration improvements:
object-group network ipv4 SRC_1 object-group network ipv4 DEST_2
10.10.1.0/24 40.40.40.32/30
host 4.5.6.7 host 2.3.4.5
! !
object-group network ipv4 SRC_2
20.20.1.0/24 object-group port PORT_1
host 7.8.9.10 eq domain
! range 1024 65535
object-group network ipv4 DEST_1 !
30.30.0.0/16 object-group port PORT_2
host 3.4.5.6 eq 80
range 0 1023
ipv4 access-list example

10 permit tcp net-group SRC_1 net-group DEST_1 port-group PORT_1
interface GigabitEthernet0/1/1/0
ipv4 access-group example ingress compress level 1
test notes/observations
security only ACLs in 4.3.1
no QoS or other applications
all ACLs on a given NPU must have same compression level
for *very* large ACLs, it takes 10-15 seconds to commit the changes. for
normal sized ACLs its not more than a couple of seconds.
PPS performance decreases as compression level increases
Weve taken very large infra ACLs from real use cases and able to fit 2.7M
ACEs into 62k TCAM entries
Compression levels
There are three available compression levels for a scaled ACL. (level 2 is not
used/implemented at present on the asr9k...)
level 0 simply expands the object groups and dumps into TCAM (cross product)
identical performance to legacy ACL
Benefit: more convenient configuration
level 1 compresses only the source prefix object-groups
smallest performance hit, but still very high scale
level 3 compresses both SRC/DEST, pfx and port groups
higher performance reduction, but wicked-crazy-massive scale improvements
General recommendation: use least compression that fits.
more flexibility to trade performance vs. scale vs. cost
do NOT forget that SE cards have much larger TCAMs than TR cards!!!
Scaled ACL : counters
In the hardware, each TCAM entry points at a counter.
Regardless of legacy vs. object-group config, each configured ACE will have one
counter associated.
Scaled ACL allows you to combine lots and lots of rules into a single ACE, which
also becomes a single counter.
IF you need more granularity in your counters, break out a separate rule (just like
before, but with more flexibility)
Still order-dependent, so use sequence numbers...
scaled ACL commands
show pfilter-ea fea ipv4-acl <ACL> loc <loc>
shows you how many ACEs, how many TCAM entries, and TCAM entries
per ACE (must be applied to see)
show pfilter-ea fea summary loc <loc>
shows how many total ACEs/TCAM entries/stats counters are used on the
linecard (per NP, where NP=chan#)
show access-lists ipv4 <acl> hardw ing resource-usage LOC
shows compiled ACL hardware stats (TCAM, compression, etc)
show controller np struct SACL-PREFIX summary loc 0/0/cPU0
shows prefix usage for compressed tables
Side note: use new apply-groups to manage config
group MY_ACL_INTF
interface 'TenGigE0/[02]/0/[0-2]'
ipv4 access-group example1-compressed ingress compress level 1
!
end-group
group ospf-defaults
router ospf '1'
area '0'
interface 'TenGigE.*'
network point-to-point
dead-interval 8
hello-interval 2
end-group
Performance PPS impact of using scaled ACL
No ACL no features: 44Mpps
Uncompressed ACL: Input or Output ACL cost about ~10%
Level 1 compression: Input ACL or Output ACL only cost about ~20%
Level 3 compression: Input ACL or Output ACL cost about ~50%
Performance degradation is because of tree lookup in search memory

Remember that deny ACE cost less performance (packet is gone from pipeline)
Well talk more about that later
Non hybrid, tree based ACL differ in performance where you match in the ACL,
ASR9000 does NOT suffer from that (TCAM!)
Disclaimer: These are indicational numbers from benchmarking only, specific to release and subject to variation
Performance overview
ASR9k: combination of short prefix trees and
O(1) TCAM lookups. Very consistent
performance based on compression levels.
asr9k (level 0)
asr9k (level 1)
ACL performance (MPPS)
asr9k (level 3)
Competitor X
Tree based only
10^2 10^4 10^5 10^7 10^9

# of expanded rules in the ACL
Few words IOS-XR and IOS differences
What are key differences between IOS and XR
Micro kernel vs Monolithic
Process crashes are confined in XR
Ability to patch individual processes (via SMUs) (SMU manager tool!)
SNMP architectural differences (caching)
IPC (inter process communications)
Memory management and CPU utilization
EVC model (as opposed to IEEE in IOS)
Routing protocol behavioral differences
E.g. RPL instead of route-maps
E.g. BGP no sync and deterministic MED is always on things like that
Task based command author
Two stage commit
Google ASR9000 ios to xr migration guide
SMU Management Architecture
Customer SMU Manager Cisco
Intranet
Internet
www.cisco.com
Secure
Automated
Cisco Tools
Connection
SW management
capabilities
Auto Discovery PIMS
Multi Node Release Ops
Recommendations SMU Tool
Analysis and Optimization
MicroKernel instead of Monolithic
Complete Micro Kernel allowing for individual process restarts
No runaway processes
One misbehaving process will not affect another
Patchable at the individual process level
Process isolation
BGP OSPF BGP RIP OSPF BGP OSPF
Process restart EIGRP ISIS LDP VPN ISIS EIGRP ISIS
Green areas RIP VPN Telnet RIP VPN

SSH
cannot restart
Preemptive multitasking SSH Telnet Server SSH Telnet
Server Server
LDP ACLs ACLs LDP ACLs
IPv4 Forwarding IPv4 Forwarding IPv4 Forwarding

TCP/IP Drivers
TCP/IP Drivers TCP/IP Drivers
Timers Scheduler Timers Scheduler Timers Scheduler
Monolithic Kernel Microkernel

IOS BSD based routers IOS XR
Virtual memory spaces and allocation
Each process has its own dedicated memory space
Mapped to real HW addresses invisible to process
One process cannot corrupt anothers memory
Process can only access virtual space
In IOS all processes shared same virtual space 0x000000
0x100000
No more SYS-MEMDUMP! 0x200000
Comm. between procs via controlled APIs 0x300000

0x400000
0x500000
0x600000
1 0x00000 0x700000
OSPF 2 0x10000
3 0x800000
0x20000
0x900000
0xa00000
App1 App2 R SNMP architecture
X
snmpd Concurrent Requests to
4.2+ Behavior multiple MIB daemons
- Q per MIBD
engine
Processing
Serialized PDU
Processing
Serialized PDU
Q dispatcher Q
MIBD
interface MIBD Entity
caching FOO-MIB dll
GSP (different node) sysdb LWM (same node)

HW HW
ASR 9000 Flexible Ethernet SW Infrastructure
(EVC SW Infrastructure)
L3
EFP (Ethernet
Routing
Flow Point) or
sub-interface EoMPLS PW
Bridging
VPLS
EoMPLS PW
Flexible VLAN tag
classification P2P VPWS
Flexible Ethertype EoMPLS PW

(.1Q, QinQ, .1ad)
Flexible VLAN tag X Bridging Multipoint bridging
manipulation
P2P local
connect
1
Flexible service mapping and multiplexing

2
L2 and L3, P2P and MP services concurrently on the same port
Flexible Service L2VPN P2P
EFP configuration example
Interface gig 0/0/0/1.101 l2transport L2VPN P2P service configuration example
encapsulation dot1q 101 second 10 l2vpn
rewrite ingress pop 2 Symmetric xconnect group cisco
p2p service1 local connect
Interface gig 0/0/0/2.101 l2transport interface gig 0/0/0/1.101
encapsulation dot1q 101 interface gig 0/0/0/2.101
rewrite ingress pop 1 Symmetric p2p service2 VPWS
interface gig 0/0/0/3.101
Interface gig 0/0/0/3.101 l2transport neighbor 1.1.1.1 pw-id 22
encapsulation dot1q 102 p2p service3 PW stitching
rewrite ingress push dot1q 100 Symmetric neighbor 2.2.2.2 pw-id 100
neighbor 3.3.3.3 pw-id 101
Two logical ports (EFP or PW) form one EVC (Ethernet virtual circuit)
No MAC learning/forwarding involved
IOS-XR vs. IOS EVC Comparison
Common part
Both share the same EVC SW infrastructure
Feature parity for the flexible VLAN tag classification, VLAN tag rewrite and service mapping
7600 IOS
VLAN tag classification, rewrite, service mapping are all done on the port level (with some
exceptions), which is classic IOS CLI
Introduced service instance configuration mode for better L2VPN scale
Legacy switchport feature support in parallel (but cant co-exist with EVC on the same port)
IEEE trunks
Interface VLAN
ASR 9000 IOS-XR

De-couple port level and service configuration. VLAN tag classification and rewrite are done at
port level. L2VPN services are configured at l2vpn module
Uniform sub-interface CLI for both L2 and L3 service, no additional service instance structure
Common Infrastructure for native L2 and MPLS based L2VPN service
EFP based access model.
Bridge domain per vlan
BVIBRKSPG-2904 2013 Cisco and/or its affiliates. All rights reserved. Cisco Public 172
EVC Configuration Comparison (1) L2VPN P2P service
ASR 9000 7600
interface GigabitEthernet4/1/0
Local EFP configuration under interface
service instance 101 ethernet
encapsulation dot1q 101 second 10
Connect Including VLAN tag encapsulation, tag rewrite,

Qo/ACL features, etc
rewrite ingress tag pop 2 Symmetric
Interface gig 0/0/0/1.101 l2transport encapsulation dot1q 100
encapsulation dot1q 101 second 10 rewrite ingress tag pop 1 Symmetric
connect eline-101 GigabitEthernet4/1/0 101 GigabitEthernet4/1/1 100
Interface gig 0/0/0/2.101 l2transport
EoMPLS Service configuration under l2vpn service instance 11 ethernet
encapsulation dot1q 101 second-dot1q 60-70
l2vpn xconnect 10.0.0.3 101 encapsulation mpls
xconnect group cisco
p2p service1 local connect
l2 vfi tac-training point-to-point
PW interface gig 0/0/0/2.101
p2p service2 EoMPLS
neighbor 10.0.2.3 3001 encapsulation mpls
neighbor 10.0.2.2 3000 encapsulation mpls
stitching neighbor 1.1.1.1 pw-id 22
p2p service3 PW stitching
[note] require BGP configuration if its for inter-AS
Flexible Service L2VPN Multi-Point
L2VPN MP service configuration example
EFP configuration example l2vpn

bridge group cisco
bridge-domain domain1 local bridging
Interface gig 0/0/0/1.101 l2transport Interface gig 0/0/0/1.101
encapsulation dot1q 101 split-horizon group no bridging among same SHG
rewrite ingress pop 1 Symmetric Interface gig 0/0/0/2.101
split-horizon group
bridge-domain domain2 vpls
Interface gig 0/0/0/1.101
rewrite ingress pop 1 Symmetric Interface gig 0/0/0/2.101
vfi cisco
Interface gig 0/0/0/3.101 l2transport neighbor 192.0.0.1 pw-id 100
encapsulation dot1q 102 neighbor 192.0.0.2 pw-id 100
rewrite ingress push dot1q 100 Symmetric
bridge-domain domain3 h-vpls
More than two logical ports (EFP or PW) belong to the Interface gig 0/0/0/1.101
same bridge domain neighbor 192.0.0.3 pw-id 100 spoke PW
MAC learning/forwarding involved vfi cisco core PWs
neighbor 192.0.0.1 pw-id 100 core PW
Bridge-domain is global significant, VLAN ID is local neighbor 192.0.0.2 pw-id 100
port scope
CLI Comparison (4) SVI
ASR 9000 IRB/BVI* Example
(equivalent to 7600 SVI feature) 7600 SVI example
encapsulation dot1q 50 interface gig 1/2
rewrite ingress tag pop 1 Symmetric switchport
Interface gig 0/0/0/2.50 l2transport switchport mode trunk
encapsulation dot1q 50 switchport trunk allow vlan 50-1000
l2vpn encapsulation dot1q 50
bridge group cisco rewrite ingress tap pop 1 sym
bridge-domain domain50 bridge-domain 50
Interface gig 0/0/0/1.50
Interface gig 0/0/0/2.50 Interface vlan 50
routed interface bvi 20 ip address 1.1.1.1 255.255.255.0
Interface bvi 20 *QOS policing and ACL supported on BVI starting XR43.
ipv4 address 1.1.1.1 255.255.255.0 (features replicated to all npus with EFPs in that BD!
Multiple Services on the same port example
Access port core interface, L2 trunk or L3 MPLS
interface gig 0/0/0/1.1 l2transport

encapsulation dot1q 20 second-dot1q 10
rewrite ingress tag pop 1 sym
l2vp
Local Bridging
bridge group cisco
bridge-domain cisco
interface
interface gig 0/0/0/1.2 l2transport

encapsulation dot1q 11-100 E-LINE (VPWS)
rewrite ingress tag push dot1q 101
L2vpn
xconnecct group
p2p eline-1

Local encapsulation dot1q 101 second-dot1q 10 E-LAN (VPLS)
rewrite ingress tag translate 2-to-1 100
connect l2vpn bridge group vpls
bridge-domain vpls
vfi APPLE
l2vpn
xconnect group LocalConnect Local
p2p someone
connect
Ethernet Flow Point interface GigabitEthernet0/0/0/1.6

encapsulation dot1q 200 second 200
ipv4 address 1.1.1.1 255.255.255.0
L3 service
MAC Learning Learn from Data Plane Flooding
DMAC unknown/broadcast
Precondition: SMAC unknown, DMAC

unknown/broadcast Linecard 2
1. Frame with unknown SMAC & DMAC
address enters the system on LC1 into Data Plane
BD1 A
5
P2
BD
1
2. MAC lookup, MAC table on LC1 is Linecard 1
updated with SMAC (ingress data- P3
plane learning) 4 SA=A, DA=B 6
P4
Data Plane
3. Since DMAC is unknown, frame is A
2
P2
flooded towards linecards which BD
participate in BD and to locally 1
attached ports
P1
4. LC2 and LC3 receive flooded frame P2
1
SA=A, DA=B 3
copy with unknown SMAC & DMAC
into BD1 Linecard 3
5. MAC lookup, MAC table on LC2, LC3 Data Plane

is updated with SMAC (egress data- 5
plane learning) BD
A P2
1
6. Since DMAC is unknown, frame is
flooded towards local bridge ports on P5
BD1 4 SA=A, DA=B 6
P6
MAC withdrawal / flush
A Flush is done on a per port basis, but with a mac wildcard.
This means that a vpls ldp mac withdrawal message is sent to flush basically
all macs in the Bridge domain.
This means that the Bridge domain will start to flood for a little bit, but this is no
problem considering we have hardware learning.
Pay attention to the MAC_MOVE np counter

MAC_NOTIFY is an update for learning a new mac. The npu will generate and
flood a mac-notify to all npus in the system (regardless whether they have a
bridge-domain or not)
VLAN rewrite Considerations
VLAN Tags and Pseudowires
EVC Context Service Delimiters
Virtual
PE PW Context Port-based PE
Service Delimiters Service
PW PW
Forwarder Forwarder
S-I S-I
Pseudowire
(PW)
Ingress Push Dummy VLAN (Type 4) Pop Dummy VLAN (Type 4) Egress
Encapsulation Pass-through (Type 5) Pass-through (Type 5)
Encapsulation
Adjustment Adjustment
EVC Encapsulation Adjustment is independent of negotiated Pseudowire (PW) Type; PW type dictates
VLAN adjustment in PW Forwarder only
For Ethernet PW (Type 5), frames pass through PW Forwarder with the Ethernet header unmodified
For VLAN PW (Type 4), the PW Forwarder adds Dummy VLAN in imposition path and rewrites that
VLAN in disposition path
Golden rule, always pop the service delimit VLAN tag regardless of the VC type
References
ASR9000/XR Feature Order of operation
ASR9000/XR Frequency Synchronization
ASR9000/XR: Understanding SNMP and troubleshooting
Cisco BGP Dynamic Route Leaking feature Interaction with Juniper
ASR9000/XR: Cluster nV-Edge guide
Using COA, Change of Authorization for Access and BNG platforms
ASR9000/XR: Local Packet Transport Services (LPTS) CoPP
ASR9000/XR: How to capture dropped or lost packets
ASR9000/XR Understanding Turboboot and initial System bring up
ASR9000/XR: The concept of a SMU and managing them
ASR9000/XR Using MST-AG (MST Access Gateway), MST and VPLS
ASR9000/XR: Loadbalancing architecture and characteristics
ASR9000/XR Netflow Architecture and overview
ASR9000 Understanding the BNG configuration (a walkthrough)
ASR9000/XR NP counters explained for up to XR4.2.1
ASR9000/XR Understanding Route scale
ASR9000/XR Understanding DHCP relay and forwarding broadcasts
ASR9000/XR: BNG deployment guide
References
ASR9000/XR: Understanding and using RPL (Route Policy Language)
ASR9000/XR What is the difference between the -p- and -px- files ?
ASR9000/XR: Migrating from IOS to IOS-XR a starting guide
ASR9000 Monitoring Power Supply Information via SNMP
ASR9000 BNG Training guide setting up PPPoE and IPoE sessions
ASR9000 BNG debugging PPPoE sessions
ASR9000/XR : Drops for unrecognized upper-level protocol error
ASR9000/XR : Understanding ethernet filter strict
ASR9000/XR Flexible VLAN matching, EVC, VLAN-Tag rewriting, IRB/BVI and defining L2 services
ASR9000/XR: How to use Port Spanning or Port Mirroring
ASR9000/XR Using Task groups and understanding Priv levels and authorization
ASR9000/XR: How to reset a lost password (password recovery on IOS-XR)
ASR9000/XR: How is CDP handled in L2 and L3 scenarios
ASR9000/XR : Understanding SSRP Session State Redundancy Protocol for IC-SSO
ASR9000/XR: Understanding MTU calculations
ASR9000/XR: Troubleshooting packet drops and understanding NP drop counters
Using Embedded Event Manager (EEM) in IOS-XR for the ASR9000 to simulate ECMP "min-links"
XR: ASR9000 MST interop with IOS/7600: VLAN pruning
Summary
So what have we discussed today
ASR9000 architecture overview
How the NPU forwarders work
How to troubleshoot the ASR9000 packet forwarding issues
Loadbalancing
Punt Path
Multicast
QOS architecture
Quick Comparison between IOS and XR
L2VPN/EVC configuration model and Mac learning
Call to Action
Visit the World of Solutions:-
Cisco Campus
Walk-in Labs
Technical Solutions Clinics
Meet the Engineer
Lunch Time Table Topics, held in the main Catering Hall
Recommended Reading: For reading material and further resources for this
session, please visit www.pearson-books.com/CLMilan2014
Complete Your Online Session Evaluation
Complete your online session
evaluation
Complete four session evaluations
and the overall conference evaluation
to receive your Cisco Live T-shirt

BRKSPG-2904 Miami PDF

Uploaded by

Copyright:

Available Formats

BRKSPG-2904 Miami PDF

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

BRKSPG-2904 Miami PDF

Uploaded by

Copyright:

Available Formats

ASR-9000/IOS-XR hardware Architecture,

QOS, EVC, IOS-XR Configuration and

Aleksandar Vidakovic CCIE #5697 Technical Leader

ASR 9001 ASR 9922

Max Capacity 770G/slot 440G/slot 440G/slot 550G/slot 1.2T/slot

Size 2RU 6RU 10RU 21RU 30RU 44RU

Power Feed Cabling from the Back

R/S/I Power Supply Voltage Current

R/S/I Power Draw Voltage Current

R/S/I Power Draw Voltage Current

Slot Max Watts

Flows exit in-order

Multicast traffic hashed based on (S,G) info to maintain flow integrity

Max Min 0 (Empty)

0-1 XBAR links to RSP0 (Trident+RSP2)

********** FIA-0 **********

********** FIA-0 **********

********** FIA-1 **********

Low priority stats (multicast)

Port statistics for xbar:0 port:1

Low priority stats (multicast)

It has both central RP and LC CPU like big chassis

5 Stages: All packets go through

L2/L3 header Performs QoS Processes Adds internal Queuing,

First identify interface in question with problem

RP/0/RSP0/CPU0:A9K-BNG#show controller np ports all loc 0/0/cpU0

NP Bridge Fia Ports

Delta between received from Fab to TX-wire should almost be 0, if not, we

Some counters have an index to a port.

Warning: A mandatory NP reset will be done after monitor to clean up.

Tue Jan 28 10:10:22 2014 -- NP0 packet

From OAM RTC: 64 byte packet, bytes[0-3] invalid!

Cleanup: Confirm NP reset now (~50ms traffic outage).

LPTS Internal FIB

LPTS enables applications to reside on any or all RPs, DRPs, or LCs

Any ICMP ANY ANY 1000 low

LC 2 IFIB TCAM HW Entries

If you can use only p2p OSPF network type

RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh lpts pifib hardware entry brief location 0/4/CPU0 | i any,23

NP Trident Typhoon Show controllers np ports all location <>

show controllers fabric show controllers fabric

Jasper Dao Skytrain Sacramento

8641D TSEC2 Punt Octopus

Show controllers np ports all loc <LC>

P4040 1G DPAA Punt Skytrain Sacramento

LSD RIB RSP CPU

AIB: Adjacency Information Base

LSD RIB RSP CPU

ARP SW FIB HW FIB

AIB: Adjacency Information Base

ARP SW FIB HW FIB

Backup TE Adj OIF

LSD RIB RSP CPU

RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#sh route 10.1.2.0/24

Routing entry for 10.1.2.0/24

Entry missing or not what you have expected?LC CPU

RX H/W Result on NP:0 [Adj ptr:0x18 (BE)]:

Ifname(1): GigabitEthernet0_7_0_2, ifh: 0x12000140 :

RP/0/RSP0/CPU0:EAST-CORE-ASR9K-1#show uidb index location 0/7/CPU0 gig0/7/0/2

FIA-0

FIA-0

FIA-1