The Association of System

Performance Professionals

The Computer Measurement Group, commonly called CMG, is a not for profit, worldwide organization of data processing professionals committed to the
measurement and management of computer systems. CMG members are primarily concerned with performance evaluation of existing systems to maximize
performance (eg. response time, throughput, etc.) and with capacity management where planned enhancements to existing systems or the design of new
systems are evaluated to find the necessary resources required to provide adequate performance at a reasonable cost.

This paper was originally published in the Proceedings of the Computer Measurement Group’s 2004 International Conference.

For more information on CMG please visit http://www.cmg.org

Copyright Notice and License

Copyright 2004 by The Computer Measurement Group, Inc. All Rights Reserved. Published by The Computer Measurement Group, Inc. (CMG), a non-profit
Illinois membership corporation. Permission to reprint in whole or in any part may be granted for educational and scientific purposes upon written application to
the Editor, CMG Headquarters, 151 Fries Mill Road, Suite 104, Turnersville , NJ 08012.

BY DOWNLOADING THIS PUBLICATION, YOU ACKNOWLEDGE THAT YOU HAVE READ, UNDERSTOOD AND AGREE TO BE BOUND BY THE
FOLLOWING TERMS AND CONDITIONS:

License: CMG hereby grants you a nonexclusive, nontransferable right to download this publication from the CMG Web site for personal use on a single
computer owned, leased or otherwise controlled by you. In the event that the computer becomes dysfunctional, such that you are unable to access the
publication, you may transfer the publication to another single computer, provided that it is removed from the computer from which it is transferred and its use
on the replacement computer otherwise complies with the terms of this Copyright Notice and License.

Concurrent use on two or more computers or on a network is not allowed.

Copyright: No part of this publication or electronic file may be reproduced or transmitted in any form to anyone else, including transmittal by e-mail, by file
transfer protocol (FTP), or by being made part of a network-accessible system, without the prior written permission of CMG. You may not merge, adapt,
translate, modify, rent, lease, sell, sublicense, assign or otherwise transfer the publication, or remove any proprietary notice or label appearing on the
publication.

Disclaimer; Limitation of Liability: The ideas and concepts set forth in this publication are solely those of the respective authors, and not of CMG, and CMG
does not endorse, approve, guarantee or otherwise certify any such ideas or concepts in any application or usage. CMG assumes no responsibility or liability
in connection with the use or misuse of the publication or electronic file. CMG makes no warranty or representation that the electronic file will be free from
errors, viruses, worms or other elements or codes that manifest contaminating or destructive properties, and it expressly disclaims liability arising from such
errors, elements or codes.

General: CMG reserves the right to terminate this Agreement immediately upon discovery of violation of any of its terms.
 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

WORKLOAD CHARACTERIZATION FOR WEBSPHERE APPLICATION

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
SERVER

Yefim Somin
BMC Software, Inc.
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

Abstract

WebSphere Application Server is a complex Java-based

environment widely used as a business tier for critical
applications accessed through the web. It is important to
understand what resources are consumed within
WebSphere by different types of business activities. Using
OS level process information or other non-process statistics
presents certain problems. The paper discusses benefits
and problems inherent in various approaches.

1. Introduction components – Servlets, JAVA Server Pages

(JSP’s), and Enterprise Java Beans are
deployed to carry out functional tasks.
The server at the end of the network is the
critical element in the delivery of the content to
Naturally performance management of such an
the user. It has to be deployed, configured and
environment is a complicated expert task. Many
utilized in the optimal way to provide the required
materials have appeared over the last few years
quality of service while operating efficiently. It is
to help measure, configure, tune and tweak this
therefore critical to measure the relevant aspects
environment. Both IBM (see for instance
of the functioning of the server, most importantly
[JOIN02] on performance of Java Web Sites)
in relation to the workload imposed on the
and outside experts (e.g., [NEAT04] on
server. Correct measurements are the
WebSphere performance in particular) have
foundation of correct optimization and
pitched in. However, capacity planning for
achievement of service level goals.
WebSphere and especially its prerequisite –
workload characterization – have not been
WebSphere Application Server is a leading
addressed in literature and have been left to the
environment for providing web-based content
few practicing experts knowledgeable in this
and application services for all lines of
specific area. The goal of this paper is to provide
businesses. Like most such environments, it’s
information on the current state of affairs for
based on Java language and other related
WebSphere workload characterization and
concepts and constructs. Unlike traditional
capacity planning. Relevant architectural
Mainframe environments, this context presents
features of this environment, available data and
extra challenges. It is multi-layered with abstract
approaches are covered, while the existing
objects at several virtualized levels. There are
roadblocks are also pointed out.
OS objects – global services, processes,
threads. Java Virtual Machine (JVM) is the next
level environment in which application services
operate. WebSphere’s own organizational 2. Workloads and WebSphere
structures such as Cells/Domains, Containers Application Server
are based on JVM’s. Finally, low level

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

Workload characterization is a process providing Currently, there are two main versions of

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
information on how computing resources are WebSphere deployed in the field: WebSphere 4
used for distinct business tasks or by different and WebSphere 5. They differ in management
groups of users. Knowledge of workloads makes utilities, partitioning hierarchies and
it possible to address critical performance nomenclature. A good comparison from the point
management and capacity planning questions, of view of a systems administrator is given in
such as: [NEAT04]. For the purposes of this paper, most
of these differences are not important, and the
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

- what applications are responsible for following discussion applies to both. It is duly
current resource use and possibly noted when information specific to WebSphere 5
immediate performance problems is given. This version is expected to dominate in
(account balancing vs. online queries, or the coming future. The whole discussion is
customer browsing vs. supplier dedicated to WebSphere on distributed systems.
requests) The mainframe environment is significantly
- immediate and longer term planning: different.
o current modifications/upgrades
to solve service problems 3.2 WebSphere Environment Tiers
o capacity necessary to
accommodate business growth WebSphere environment has a multi-tier
configuration. There could be up to 4 tiers. The
Workloads have attracted a lot of attention at tiers are logical and could be distributed over a
CMG’03 as a critical element in the performance multi-node configuration either vertically
analysis practice. For an extensive coverage of (different tiers in different physical locations) or
workloads as a science and an art see [KAMI03]. horizontally (peer components of the same tier
The first step in analyzing WebSphere workloads could run on parallel nodes). The tiers are:
is to determine what resources are used by
WebSphere-related activities, as distinct from • Web Clients
other tasks present in the systems in question. • HTTP Servers
However, it is not sufficient. Today’s computing • Application Servers
environment management is business user- • Database Servers
oriented, hence, specific application analysis is
required. This, in turn, demands knowledge of In a variety of configurations available from IBM,
what applications and preferably their some tiers can collapse into others, e.g., an
components consume resources in which part of HTTP Server can be embedded into an
a distributed environment. To continue the Application Server.
discussion in more technical detail, the
architecture of the environment needs to be 3.2.1 Web Client Tier
addressed.
Clients are remote and connected to the
services using Web Browsers. For the purposes
Workload characterization consists of breaking
of workload groupings, identities of the clients
down resource utilization in a computing
are difficult if not impossible to trace through all
environment and assigning the work to
the tiers of the environment. This task is beyond
appropriate categories. While determination of
the scope of the present paper.
what constitutes a workload in a particular
environment is in the eye of the analyst, it also
depends on the kind of statistics available for
3.2.2 HTTP Server Tier
this purpose.

HTTP Servers handle internet traffic from Web

3. WebSphere Application Server Clients as well as communication with
Architecture Application Servers. There are numerous
approaches to the setup and configuration of
3.1 General Notes these servers, common to a variety of web

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

environments. They are covered in literature and administrative grouping of WebSphere runtime
are not addressed in detail here. Generally, for a components, usually over a distributed

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
given amount of end-user requests, the resource configuration. Node is a subset of a Cell located
consumption by HTTP Services is significantly on one physical node (another terminology
smaller than for the application specific overload). One WAS runs within one JVM (Java
processing on an Application Server. When Virtual Machine), a standard environment for an
carried out separately, HTTP processing is done interrelated set of Java programs to run in and
by processes with well-known command names. use a set of Java services. On distributed
systems, one JVM is implemented within one OS
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

process. This is an important attribute to

3.2.3 Application Server Tier consider when it comes to workload
characterization for WebSphere. To enable
WebSphere Application Server (WAS) is the parallel handling of multiple requests as well as
main focus of this discussion. It implements all service and support functions, Java
business and presentation processing, including implementation relies on Java threads, which in
communication with the database. It could be turn are mapped onto OS level threads.
deployed in a variety of configurations: Statistics on Java thread pools are available for
collection, while obtaining statistics on individual
• Multiple WAS’s on a computing node threads is a much tougher technical problem to
• WAS’s distributed over multiple nodes solve.
• WAS functions split among JVM’s (Java
Virtual Machines) A WAS server includes a Web Container and a
EJB Container (Fig.2). A Web Container
Since WAS workload characterization is the includes presentation components:
target of this paper, a more detailed discussion
of its structure is presented in Section 3.3. • JSP’s
• Servlets
• Static content
3.2.4 Configuration Example
An EJB Container includes EJB’s (Enterprise
Java Beans) responsible for business logic and
An example of a WebSphere configuration is communication with the database. While these
presented on Figure 1. It is a small subset of a groupings and objects do not by themselves
configuration that may exist at a large determine the workload breakdown, they are
installation. It contains Web Clients, typically involved in workload resource consumption as
remote, 2 HTTP Servers receiving client described in a later section on threads.
requests and capable of directing them to any of
the 4 WAS Servers located on 2 physical nodes.
A Database Server is connected to all WAS
Servers.

3.3 WAS Architecture - details

A WAS instance in WebSphere terminology is

called simply Server - a terminology overload to
be aware of. A default personal label for a server
at its creation offered by WebSphere is
“server1”. See for instance Fig.1 with two
instances (servers) configured per node, and
with instance names Server1, Server2, and so
on. Server in WebSphere 5 is the lowest level of
management configuration hierarchy: Cell –
Node – Server. Here Cell is a high level

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
Web Web Web Web
Client Client Client Client
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

HTTP Server HTTP Server

WebSphere Server Node A WebSphere Server Node B

WebSphere Application Server WebSphere Application Server

WAS “Server1” WAS “Server3”

WebSphere Application Server WebSphere Application Server

WAS “Server2” WAS “Server4”

Database
Server

Figure 1. WebSphere Architecture Overview

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

WAS “Server1”

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
Web Container

JSP
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

JSP JSP

Servlet

EJB Container

EJB
EJB

EJB
EJB

Figure 2. WAS Containers and Objects

4. WebSphere Workloads 4.2 Subapplications

Enterprise Application may be further subdivided

4.1 Enterprise Applications into subapplications. Petstore for instance
consists of:
A natural unit for workload breakdown in a
WebSphere environment is Enterprise • Front End (Pet Gallery)
Application. Take for example a sample • Order Processing Center
application provided with WebSphere – Petstore. • Supplier Application
It represents a typical retail web-based operation
where a customer access scenario can be For a more detailed description of Petstore and
emulated. Another sample application provided its benchmarking see for instance [JUSE03].
with the software is PlantsByWebsphere. In a A big Enterprise Application may have:
more typical corporate environment,
departments like Finance, Sales, etc. may have • Account queries
their own Enterprise Applications. • Receivables
• Etc.

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

rver/properties:/usr/WebSphere/AppSe
rver/lib/bootstrap.jar:/usr/WebSpher
e/AppServer/lib/j2ee.jar:/usr/WebSph

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
4.3 Workload allocation by server
ere/AppServer/lib/lmproxy.jar:/usr/W
ebSphere/AppServer/lib/urlprotocols.
In many environments a WAS server is assigned
jar -Xms50m -Xmx256m -Dws.ext.dirs=/
to one particular application. Indeed, in large usr/WebSphere/AppServer/java/lib:/us
environments one application may be spread r/WebSphere/AppServer/classes:/usr/W
over a number of WAS servers. In a simplified ebSphere/AppServer/classes:/usr/WebS
example on Fig.3 the application assignment is phere/AppServer/lib:/usr/WebSphere/A
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

as follows: ppServer/lib/ext:/usr/WebSphere/AppS
erver/web/help:/usr/WebSphere/AppSer
• Petstore: WAS Server1, PID 18284 ver/deploytool/itp/plugins/com.ibm.e
• FlowerShop: WAS Server2, PID 19888 tools.ejbdeploy/runtime -Dserver.roo
t=/usr/WebSphere/AppServer -Ddb2j.sy
As has already been mentioned, each WAS stem.home=/usr/WebSphere/AppServer/c
loudscape -Duser.install.root=/usr/W
server is implemented as a JVM within an OS
ebSphere/AppServer -Dcom.ibm.itp.loc
process. It is therefore sufficient to find which ation=/usr/WebSphere/AppServer/bin -
process implements which application to have a Dwas.install.root=/usr/WebSphere/App
set of metrics reflecting resource consumption Server -Djava.security.auth.login.co
for that application. This turns out to be a less nfig=/usr/WebSphere/AppServer/proper
than trivial task. PID’s (Process ID numbers) for ties/wsjaas.conf -Djava.security.pol
the example above are included for information icy=/usr/WebSphere/AppServer/propert
purposes, but in fact the connection between an ies/server.policy com.ibm.ws.bootstr
application and a corresponding OS process is ap.WSLauncher com.ibm.ws.runtime.WsS
not readily available. The problem with erver /usr/WebSphere/AppServer/confi
establishing such a connection is that the visible g aix-pmw-qa-s2Network aix-pmw-qa-s2
short command name for each of the processes server1
involved is simply java.
The following observations can be made:
Can a full command name help in identifying
processes with applications? • WebSphere command names are very long
– this one is over 1KB in length
4.4 Using full command names for workload • The short command name is still java (see
identification the last token of the leading path string)
• It seems to contain strings identifying it as
The following two questions need to be WebSphere-related, namely WebSphere (in
answered to determine if full command names other cases WebAS may appear)
could be used for workload identification: • It contains strings matching WebSphere 5
configuration objects: aix-pmw-qa-
• Does the full command name contain the s2Network aix-pmw-qa-s2 server1 happen
necessary info? to be the names of Cell, Node, and Server
• Can this info actually be obtained? for the examined configuration.

To address both of these questions let us look at Thus, the answer to the question about the
a typical example of a full command name for a presence of useful information seems to be yes.
process implementing a WAS server: In fact, by using a pattern-matching algorithm,
one can progressively narrow the granularity of
/usr/WebSphere/AppServer/java/bin/ja the workload characterization:
va -Djava.net.preferIPv4Stack=true -
Dwas.status.socket=57972 –Xbootclass • Total Java work is obtained by filtering for
path/p:/usr/WebSphere/AppServer/java java
/jre/lib/ext/ibmorb.jar:/usr/WebSphe
• Overall WebSphere work is separated by
re/AppServer/java/jre/lib/ext/ibmext
.jar -classpath /usr/WebSphere/AppSe filtering for WebSphere or WebAS
rver/properties:/usr/WebSphere/AppSe • A specific WAS server can be identified by
filtering for its configured name

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

WAS Server1

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
Petstore
Command: java
PID: 18284
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

WAS Server2

FlowerShop
Command: java
PID: 19888

Figure 3. Workloads by WAS Server

The potential problem is that the strings listed candidate. However, unless it’s used with
above are included in the full command name appropriate switches, the length of the command
with a diminishing degree of certainty. The word name string is truncated far short of the size
java is always present in Java processes. The needed. It is recommended, therefore, to use
strings WebSphere or WebAS are included something like
based on the installation directory path, which is
set by default, but could theoretically be /usr/ucb/ps -auxww
changed. Finally, although Cell, Node, and
Server names have been found in all although the specifics of the switches may be
environments we observed, there is no platform dependent.
documentation or statement from IBM that
certifies or guarantees this fact. On a Windows platform, there is no user level
utility that delivers a full command name string.
Let us assume, however, that the appropriate
strings are present in the full command name. These strings are kept in special kernel
Can we obtain this extremely long command structures, and appropriately written tools can
name? deliver them to the user. If such tools can also
facilitate filtering based on specified patterns, all
One approach would be to use standard OS the better.
utilities. Under UNIX, ps command is the prime

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

It should be noted, that WebSphere 5 • A thread is dynamically assigned to do a

programmatic interface allows determination of chunk of work (e.g., methods of a specific

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
PID for a specific server and thus can tie a EJB)
process to the server and hence to an • After completing a chunk of work a thread
application. The catch is that a tool employing can be idle for a while and then be assigned
this interface and conveying this information to another chunk of work (e.g., a Servlet)
the user needs to be engaged.
To track and attribute work to workloads,
4.5 Workload Characterization based on OS resource consumption must be collected from
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

processes threads and tied with different components in

turn related to applications and subapplications
The attributes described in the previous section of interest. Therein lies the difficulty.
allow workload characterization based on OS
processes. The system administrator must know 4.7 Fine granularity collection – present and
which servers were assigned to which future
applications. OS processes statistics under
distributed systems include metrics representing There are currently interfaces to obtain statistics
directly or indirectly: from WebSphere (JVM PI, JMX). When it comes
to resource consumption statistics collection,
• CPU consumption however, these interfaces impose a very high
• I/O load overhead. In fact, they could only be turned on in
• Memory a lab environment for evaluation and
• Other use of system resources development purposes. To use them in a
production setting would defeat the purpose of
Resource counts for each process are assigned improving performance.
to an appropriate workload and summed
according to the specifics or a particular metric. Various approaches have been mentioned as
The resulting consumptions buckets can then be potential solutions to the problem of fine
used for reporting or capacity planning. granularity collection:

4.6 Applications sharing servers • Bytecode method – a technique of inserting

compiled Java code in the appropriate
The previous discussion was under the places
assumption that each WAS Server was • AOP (Aspect-Oriented programming) – a
dedicated to one application and obtaining mechanism to have certain actions be
information on subapplications could be ignored. performed in connection with the execution
It is, however, possible to share a Server among of Java methods
more than one application, not to mention the
possibilities of interesting subapplications. Let us It should be noted, that these approaches are
consider an example on Fig.4. still research topics and the delivery of the
solution is still in the future.

Here, 2 servers are shared by 2 applications

each. Server name does not identify an 5. Workloads as foundation for
application, if only because there is more than Capacity Planning
one. What can be done in this case? Presently,
regrettably not much. A more detailed
Capacity planning is based on workloads. It
explanation has to do with the relationship
should answer questions about performance
between Java threads and WebSphere/J2EE
changes under the following “what-if” conditions:
objects.
• Workload growth – by workload
Java work within a JVM is done by threads (see
Fig.5). • Workload reallocation
• Hardware changes
• Existing threads are kept in a pool • Application changes

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

WAS Server1

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
Petstore
FlowerShop
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

Command: java
PID: 18284

WAS Server2

Autoparts
Personals

Command: java
PID: 19888

Figure 4. Applications Sharing Servers

Usually, the foundation of capacity planning

algorithms are approaches based on queueing
theory akin to those described in [LAZO84].
Commercial tools linking real measured
workloads with capacity planning / performance
modeling are available.

Find a CMG regional meeting near you at www.cmg.org/regions

 Learn the basics and latest aspects of IT Service Management at CMG's Annual Conference - www.cmg.org/conference

JSP X
EJB1 (Petstore Servlet A (Personals)
Front End) (Flowershop)
Thread 1

Buy the Latest Conference Proceedings and Find Latest Computer Performance Management 'How To' for All Platforms at www.cmg.org
Servlet Y EJB53 Servlet 28
Join over 14,000 peers - subscribe to free CMG publication, MeasureIT(tm), at www.cmg.org/subscribe

(Financials) (Escalations) (Inventory)

Thread 2

Figure 5. Allocation of threads

6. Conclusions IMPLICATIONS OF WEB SERVICES, CMG’03

Proceedings, Dallas 2003

• Workload characterization for WebSphere is [KAMI03] Ronald R. Kaminski, Time Stable

critical for capacity planning Workload Characterization Techniques, CMG’03
• Knowledge of its architecture is necessary Proceedings, Dallas 2003
for workload characterization
• With that knowledge and appropriate tools a [SOMI02] Yefim Somin and Art Rodriguez,
certain granularity of workload Workload Characterization and Capacity
characterization can be achieved Planning for DB2 Universal Database, CMG’02
• Finer granularity needs to wait for improved Proceedings, Reno 2002
instrumentation
[IBM] Useful Commands for WebSphere
Application Server, IBM On-line articles
7. Acknowledgments http://www-
106.ibm.com/developerworks/websphere/library/
techarticles/0211_rivera/rivera.html
The author thanks Mike Behne of BMC Sofware
for technical consultations and review.
[LAZO84] Edward Lazowska, John Zahorjan,
Scott Graham, and Kenneth Sevcik, Quantitative
System Performance, Prentice Hall, 1984
8. References
[NEAT04] Adam G. Neat, Maximizing
9. Brief Acronym Reference
Performance and Scalability with IBM
WebSphere, Apress, 2004
JVM – Java Virtual Machine
[JOIN02] Stacy Joines et al., Performance JSP – JAVA Server Pages
Analysis for Java Web Sites, Addison Wesley EJB – Enterprise Java Beans
Longman, 2002 WAS – WebSphere Application Server

[JUSE03] Kai S. Juse et al., PETSTORE-WS:

MEASURING THE PERFORMANCE

Find a CMG regional meeting near you at www.cmg.org/regions