IBM Global Services

Network performance
and capacity planning:
Techniques for an e-business world
e-business is about transforming
key business processes with
Internet technologies. In an
e-business world, networks are
the heart of your business.

Baseline analysis
requirements

Rapid response times. High availability. Adequate bandwidth. These are

the demands placed on your business networks. Fueled by an explosion of
e-business and remote access requirements, the performance of your network
today directly affects your customers first impression of you, and therefore
your business success, more than ever before. Effective network baseline
analysis gives you the ability to improve and maintain performance of your
existing networks as well as make informed design and purchase decisions
as you grow.
Effective network capacity planning doesnt always translate to, Buy
more bandwidth. Network requirements should be based on analytical
insights and business goals. Network baseline analysis provides essential
information using industry-standard SNMP MIB data collection techniques
and proven analysis methods. Through data gathering and analysis of
network performance trends, an enterprise view of LAN/WAN availability,
performance and capacity can be obtained. You can then effectively answer
the following performance and capacity questions:
What points in the network are directly impacting availability and
performance? Where is the congestion?
Are there times of the day when the business is vulnerable due to high use?
Is existing capacity being used effectively? Are current resources
allocated to maximize performance and minimize costs and problems?
Is the correct mix of technology employed to meet business demands?
Are network applications and protocols utilizing the network bandwidth
effectively and efficiently?

Network performance and capacity planning

Getting the full picture

To answer these questions requires an in-depth look at the network from

a number of perspectives. Lets look at the key elements that need to be
analyzed:
Traffic Characteristics
Traffic volumes and rates
Prime versus non-prime traffic rates
Traffic volumes by technology
Load Distribution/Load Balancing
Device throughput
Operational Capacity
WAN percent capacity used
LAN percent capacity used
Resource Interface Utilization
Top utilized interfaces
LAN interface (router stats) and segment (RMON stats)
WAN interface utilization
Evidence of Congestion
Packet discards
Top error interfaces
Traffic Overhead
Network traffic

Network traffic characteristics

give you a view of network
usage, helping you begin to
develop a comprehensive picture.

If you dont know where the traffic is coming from on your network, then
how do you accurately plan for it? Like cars on a highway, you need to
know how many are on the road, what type they are, where they are coming
from, and where they are going. What happens at rush hour, when your
traffic is the highest? If a road or route becomes unusable, can data still
reach its destination?
Traffic analysis demands a comprehensive look at your networks behavior.
To anticipate potential problems and prevent them, establish a networkwide
view of how much data is being transferred across the network. Determine
traffic patterns by various business periods for a given day or for the entire

page 2

Network performance and capacity planning

collection period. Information about traffic characteristics includes views of

significant network devices, reports on traffic volumes and rates that change
per unit of time for major devices. Differences in network traffic between
different network technologies can be valuable in determining network
usage practices and guide network planning activities when the information
is used effectively.
As an example, here are a few sample traffic characteristics taken from
a distributed enterprise network analysis done over a period of 6 days.
Chart 1 - Network Traffic, is an example of a network profile.

Hourly Profile Network Volume

BPS throughput
Millions

Errors
Millions
7

60

50

5
40
4
30
3
20
2
10

Ave Bits Per Sec

Feb 19 1999 2:00:00 PM

Feb 19 1999 8:00:00 AM

Feb 18 1999 8:00:00 PM

Feb 19 1999 2:00:00 AM

Feb 18 1999 2:00:00 PM

Feb 18 1999 8:00:00 AM

Feb 17 1999 8:00:00 PM

Max Bits Per Sec

Feb 18 1999 2:00:00 AM

Feb 17 1999 2:00:00 PM

Feb 17 1999 8:00:00 AM

Feb 16 1999 8:00:00 PM

Feb 17 1999 2:00:00 AM

Feb 16 1999 2:00:00 PM

Feb 16 1999 8:00:00 AM

Feb 15 1999 8:00:00 PM

Feb 16 1999 2:00:00 AM

Feb 15 1999 2:00:00 PM

Feb 15 1999 8:00:00 AM

Feb 14 1999 8:00:00 PM

Feb 15 1999 2:00:00 AM

Feb 14 1999 2:00:00 PM

Feb 14 1999 8:00:00 AM

Feb 13 1999 8:00:00 PM

Feb 14 1999 2:00:00 AM

Feb 13 1999 2:00:00 PM

Total

Chart 1: Hourly Profile Network Volume

page 3

Network performance and capacity planning

66 percent of the traffic seen during the collection period was Ethernet,
followed by 28 percent WAN and 5 percent ISDN.
The hourly network traffic profile identifies two distinct periods of
network activity: 8 a.m. - 5 p.m. Monday to Friday during the prime
shift and a daily 5 - 12 p.m. batch window.
The hourly network traffic profile identifies peaks of congestion
activity between 8 a.m. and 5:00 p.m. Monday to Friday, and peak
collision activity on the RMON-monitored Ethernet segments during
the 5 - 12 p.m. batch window.
Network load balancing and distribution

Identifying areas to implement

or modifying load balancing
for key network components
can improve availability and
performance.

To keep traffic flowing on your network highway, you will need to constantly
evaluate the load on your significant network components, especially the
routers providing Internet and intranet connections for your internal and
external customers. This will help you identify areas of vulnerability as well
as opportunities to better utilize existing resources. Using this information
will help you more effectively implement new and redundant paths throughout
your network to lower the risk of a single point of failure. Performing this
analysis will also uncover when components are either not operating, or not
configured as you had expected or planned during your design. Our experience
is that actual component utilization is often quite different from expectations.
Without this information, you may incorrectly focus your problem determination
activities on symptoms, rather than causes, of network performance concerns.
Examples of network loading statistics from our enterprise analysis include:
Of the 31 devices monitored, 4 devices contribute 60 percent of the total
network volume (Bytes) during the collection period.
DWN-Chan1 router contributes over 40 percent of the overall daily
network average of 92 Mbps, highlighting this resource as critical to
the core network infrastructure and operation of the network.
Some network device vendors provide the capability to monitor the device
CPU, memory and buffer resources. If available, this information may also
be looked at during the baseline analysis process. Incorrect or inadequate
resourcing of network devices can be a significant contributor to network
errors and device bottlenecks which constrain network traffic.

page 4

Network performance and capacity planning

Operational capacity

e-business networks require

an increased focus on WAN
capacity and performance
to support critical business
applications.

On your highways, how many lanes do you have available for traffic, what
are the speed limits, and how full are they? An understanding of your true
network speed and capacity can uncover a number of potential problem
areas and opportunities to utilize or eliminate unused resources. Historically,
the bulk of network traffic was local with workstations accessing local file or
application servers. WAN traffic has traditionally been for communications
between company sites. The network of today is WAN intensive, a result of
the higher Internet/intranet traffic capacity requirements needed to support
e-business ventures and initiatives.
Adding WAN capacity represents a real cost because of the relatively
high expense of WANs compared to LANs. New WAN usage should be
evaluated in terms of business application requirements and how they
contribute to the business. WAN capacity that has already been added to
meet anticipated needs should be reevaluated periodically. This ensures
that available capacity is being used to the planned-for degree. Chart 2
shows how capacity for a WAN connection tracks over a six-day period.

Operational WAN
Capacity BPS

Average
percent
In Capacity

Potential
Max Percent
In Capacity

Average
percent
Out Capacity

Potential
Max Percent
In Capacity

3.58

8.41

13 Feb.

46731000

2.4

14 Feb.

46731000

2.2

7.19

3.17

8.19

15 Feb.

46731000

6.33

29.8

7.5

31.05

16 Feb.

46731000

6.38

27.3

7.62

37.4

17 Feb.

46731000

6.69

27.12

7.61

31.16

18 Feb.

46731000

6.1

25.06

7.7

31.69

Chart 2: Average WAN Capacity

Effective network baseline analysis provides an accurate picture of the

relationship between available versus utilized capacity for the network.
This evaluation includes links to network vendors where the consumption
of available capacity should be evaluated 24 hours a day for 7 days to
evaluate trends. Significant cost savings can be uncovered in purchased
capacity that is either unused or may be reallocated to a link with greater
business requirements.

page 5

Network performance and capacity planning

Analysis of network capacity should include a percent of network capacity

being used, as well as a breakdown of the capacity being used on both
the LAN and WAN. In addition, it is useful to look at WAN capacity use
across diverse WAN technologies, such as frame relay or ISDN, where the
unique characteristics of the technology circuit should be reviewed by an
experienced consultant.
Our enterprise network analysis example shows the following operational
capacity:
WAN operational capacity is 46.7 Mbps for the monitored interfaces.
Maximum average inbound utilization is 6.69 percent, with a potential
maximum inbound utilization of 29.8 percent on Feb. 15. Maximum
average outbound utilization is 7.70 percent, with a potential maximum
outbound utilization of 37.4 percent on Feb. 16.
Ethernet operational capacity is 5,050 Mbps for the monitored interfaces.
Maximum average utilization is .28 percent, with a potential maximum
utilization of 1.23 percent on Feb. 16.
Of the monitored devices, router resource DWN-Chan7, with operation
capacity of 20,256 Mbps, had the highest average capacity. Capacity
used was at 13.7 percent on Feb. 16, with a maximum capacity of
72.5 percent on Feb. 16.
Overall, operational network capacity is plentiful.
Network interface utilization

Utilization analysis can

give you the performance
improvements you need
to meet business demands
without increasing your
infrastructure expense.

How busy are your highway interchanges, points where traffic leaves or enters
a network segment? Network utilization for any segment or link in the network
is determined by calculating the relationship between the volume of traffic
and the speed of the segment link. Baseline network analysis requires
characterizing the utilization of network device interfaces. Knowing which
interfaces have the highest utilization, when high utilization occurs, and if
the utilization pattern regularly exceed some identified threshold may identify
a need for more (or less) capacity. Highly utilized interfaces can also represent
potential bottlenecks. Underutilized network segments may represent
potential cost savings. Balancing traffic between over- and underutilized
segments can dramatically improve performance for your customers.

page 6

Network performance and capacity planning

Network utilization analysis should include information identifying the

most highly used segments in the enterprise network. It is also important
to characterize the direction of the incoming versus outgoing traffic. This
identifies which traffic is affecting performance and helps to identify
the source so it can be addressed. Analysis should include the average
utilization, top interfaces by maximum utilization and time periods
when send/receive interfaces exceeded a threshold. Frame-relay circuit
information must be analyzed to identify top frame-relay interfaces by
average percent sent/received utilization and frame-relay circuits with
average utilization over committed information rate (CIR).
Our enterprise resource interface utilization analysis shows:
10 WAN circuits report 60 percent or greater receive or sent utilization
at minimum 10 percent of time during prime-shift hours.
4 ISDN circuits report 60 percent or greater receive or sent utilization
at minimum 10 percent of time during prime-shift hours.
4 10Mbps Ethernet interfaces report peak 15-minute intervals of greater
than 30 percent utilization during prime-shift hours.
4 10Mbps Ethernet segments report peak 15-minute intervals of greater
than 30 percent utilization during prime-shift hours.
Network congestion

Analysis of congestion
conditions allows you to
improve response times
and availability.

Is the data on your network highway being delayed or not even reaching
its expected destination on the first trip? When data packets generated
by an application do not reach their destination due to congestion, this
information must be retransmitted. In addition, network error traffic may
be generated to indicate the source of the delivery failure. The additional
traffic associated with resending the information as well as the error
traffic can worsen network congestion and may cause your customers to
experience a delay or outage. If network congestion is a rare occurrence,
it is insignificant. If its a common occurrence, there may be a significant
impact on business operations.
Analysis of network congestion includes error packet activity reports
for the network as a whole and critical network devices specifically. The

page 7

Network performance and capacity planning

top 25 with high error activity should be identified and examined in more
detail to identify the cause. Any interface with errors greater than 10 packets
per second should also be analyzed in greater detail. Reports on discarded
packets for the network as a whole as well as critical devices can also reveal
problem areas. Any discard is an undesirable event, but it is a normal
adaptive mechanism during peak network traffic events. Occasional discard
events are insignificant, but they must be analyzed in terms of network
traffic characteristics in general. This is where the consultant experienced
in many varieties of network implementations depends on professional
skill to make the correct judgment. Chart 3 a high-level summary of traffic
activity that shows which network technology is experiencing high rates
of errors and discards.

Total Bytes
by Network
Technology

Ethernet

65.5%

Total Errors
by Network
Technology

Ethernet

Total Discards
by Network
Technology

45.2%

Ethernet

35.4%

Token Ring

0.5%

Token Ring

Token-Ring

SDLC

0.2%

SDLC

14.4%

SDLC

WAN

28.5%

WAN

39.5%

WAN

53.0%

ISDN

5.3%

ISDN

0.9%

ISDN

11.6%

Chart 3: Network Technology Rates of Errors and Discards

page 8

Network performance and capacity planning

The following is an example of an analysis showing congestion and error

traffic activity for an enterprise.
Three interfaces contribute over 70 percent of the total network discards.
One of these interfaces contributes 56 percent of the 70 percent.
Ethernet interface Ethernet3 on router resource CAS-Hosp1 reports
congestion activity during prime-shift hours. The hourly network profile
suggests that the Ethernet segment connected to this router interface
is congested. This interface contributes the majority of discards.
WAN circuits Serial5/4 and Serial4/1 on router resource DWN-Chan1
show congestion activity during prime-shift peaks in outbound utilization.
The discards, as a percent of total traffic during the congestion periods,
is less than 1 percent of the total traffic.
94 percent of network errors are reported by three router resources;
DWN-Chan1, DWN-Chan11, and DWN-Chan12. DWN-Chan1
contributes 76 percent of the 94 percent of total network errors.
68 percent of the total errors are reported by WAN circuit Serial4/6
on router resource DWN-Chan1. The errors were reported during two
distinct time periods: 9 - 10 a.m. and at 5 p.m. on Feb 18. This link
reports light utilization throughout the collection period.
WAN circuit Serial3 on router resource DWN-Chan11 consistently
reports error activity throughout the collection period. Error activity
as a percent of total packets is less than 1 percent of total.
Overall, network error activity is minimal and indicates little if any
detrimental impact to performance.
Worth noting is the fact that congestion may occur without high network
traffic loads. Inadequately provisioned network devices or network devices
that have been left at default values may create network congestion in an
otherwise well-provisioned environment. Network devices may not be tuned
to handle network traffic requirements. Monitoring network device statistics
and analyzing trends may uncover issues in the network device buffer
allocation. Device buffer pools may be inadequate or the device may be
provisioned with inadequate memory resources. These resources may have

page 9

Network performance and capacity planning

been appropriate when the device was installed, but are now inadequate
because they were not reviewed through periodic network baselines analysis
to keep them current. Reports on buffer allocation failures or buffer misses
that include 24-hour periods over at least 7 days are required to effectively
analyze this type of problem.

Controlling network and server

management reporting traffic
can increase available
bandwidth

Ethernet media is also subject to collision errors on shared segments. While

the switched environment has offered considerable relief from this type of
error on segments dedicated to a single workstation, there continue to be
shared Ethernet segments on some enterprise backbones.
Traffic overhead
Managing the traffic crossing your network highway for the purpose of
delivering management information is important for effective capacity
planning. Broadcast traffic often competes with business application traffic
and consumes network bandwidth. But broadcast traffic may be useful
and may significantly add to the apparent speed of operations by the LAN
user. Broadcast traffic over a WAN link should be analyzed and filtered to
manage this more costly and limited bandwidth.
Another part of overhead traffic is management traffic. SNMP generally
adds less than two percent of traffic, but there is potential for it to
grow uncontrollably if different groups within an enterprise implement
management stations which poll for this information. Controlling the
number of polling management stations is part of a larger management
strategy and should be part of the overall network capacity and
utilization plan.
Reports showing broadcast/multicast traffic at each interface allow the
source of the traffic to be localized at the most appropriate resource.
Monitoring for 24 hours a day for 7 days identifies patterns and cyclic
broadcast activity that would not be detectable with a standard
data-capturing tool. For accurate assessments, SNMP traffic should
be reported as a percent of total traffic.

page 10

Network performance and capacity planning

Summary

Your enterprises ability to be competitive and run cost effectively

depends on efficient networks that have the right capacity to run business
applications. Collecting and analyzing SNMP MIB data can provide a
quantitative analysis of network performance and capacity parameters in
a router- or switch-based network. It can be done with equipment from any
of the major network vendors. Effective analysis requires appropriate tools
and effective analysis methods.
Most important, however, is experience in SNMP MIB analysis. SNMP data
analysis is not a simple activity because of the flexibility in the definition
of the standards and the desire of network vendors to add value. Analysis of
MIB data for event-type activity differs from the analysis done when trending
large volumes of data to understand the impact on network performance and
capacity.
Investing in the right skills and performing periodic network baseline
analysis will allow you to maximize the performance, capacity, and
availability of your enterprise network.
For more information
For more information on IBM Performance Management and Capacity
Planning Services, call 1-800-426-4682 (in the US), 919-301-4141
(from outside the US) or e-mail us at [email protected].
William Nametka, the author of this paper, is a Senior Project Manager for
IBM Performance Management and Capacity Planning Services. His areas
of responsibility include network process, LAN design, technology migration
and MIB analysis.

page 11

 International Business Machines Corporation 1999

IBM Global Services
3200 Windy Hill Road
Mail Drop WG 15C/B27
Atlanta, GA 30339
U.S.A.
Produced in the United States of America
7-99
All Rights Reserved
IBM and the e-business logo are registered trademarks
of International Business Machines Corporation in the
United States and/or other countries.
Other company, product and service names may be
trademarks or service marks of others.
References in this publication to IBM products or
services do not imply that IBM intends to make them
available in all countries in which IBM operates.
IBMs Product Support Services organization in
the United States, part of IBM Global Services,
has successfully achieved registration to the
ISO 9001-1994 international quality standard.

For Position Only

G563-0239-00