BEA WebLogic Server Tuning Guide

While an administrative server can also be configured to host applications, it is recommended that the application

server in a domain (which has other managed servers besides the administrative server) be used for administrative

requirements only.

During steady state, the involvement of the administrative server is less, but in a production application, the

administrative server provides the monitoring of the WebLogic Server domain. So, for the purposes of estimating the

CPU cost, an administrative server can also be considered to contribute to 50% of the CPU cost of a managed

server instance.

5.2 SQL/MX Nowait Considerations

Performance testing with SQL/MX no-wait facilities has demonstrated that throughput decreases with more than four

connections per application server per Connection Pool. This is dependent on the application; for some applications

the inflection point might be much higher (for example, 20 connections per pool). Applications should monitor

connections being used and measure the performance at various limits. The recommended initial capacity is four.

5.3 Load Balancing

WebLogic Server performs load balancing based on home and remote object stubs (called replica aware stubs).

Each request gets load balanced based on the replica’s knowledge of possible target WebLogic Servers hosting the

EJB and routed to an instance.

The default load balancing strategies provided by WebLogic Server are limited in functionality. The default

strategies are round robin, weighted, and random-load balancing.

It is recommended that WebLogic Server applications (particularly those that have lots of DB access) use

parameterized load balancing along with DB partitioning. This will help in better throughput. More information on

parameterized load balancing is available from WebLogic Server documents on cluster load balancing.

5.4 Disabling the Application Poller

Performance testing has shown that running the WebLogic Server in development mode consumes significant CPU

resources because of the application polling done in development mode. It is recommended that WebLogic Server

instances in production environments use the production mode settings.

5.5 Java Heap Setting

It is recommended that the Java Heap for WebLogic Server instances be specified at 256 MB (after moving QIO to

KSEG2). This seems to provide a very good starting point. The –verbose:gc option of the WebLogic Server Java

command line can be used to study the frequency and length of GC operations. This data can be used to tune the

heap usage of WebLogic Server instances based on application requirements. Note that the Java VM allocates the

maximum required heap for JVM usage at startup. So, the swap space considerations for the JVM will be the

maximum Java Heap space specified plus other JVM memory requirements and memory requirements for native

components (the primary memory intensive component being JDBC/MX and the SQL/MX CLI).

To identify the swap usage of a process or the swap requirements for a CPU, use the NSKCOM command. For

example, from the OSS prompt use the following commands to identify the swap usage of all the processes or a

particular process:

/h/super : gtacl -p nskcom

NSKCOM - T5838G05 BASE (22JUL02) - Nov 5 2002