Tuning thread pools

For online HTTP requests and asynchronous JMS processing, Cúram is configured to use two distinct thread pools. Tune both pools as explained in the following sections.

To enable the system to be driven at optimum throughput, it is generally recommended that the total number of application threads are about twice the number of CPUs. This general convention assumes that the environment has fast I/O subsystems.

In WebSphere® Application Server (WAS), the WebContainer thread pool is set up to process HTTP requests and the SIBJMSRAThreadPool is set up to process JMS messages.

Use the following specification as a starting point: the total number of threads that Cúram uses in the application server can be set to twice the number of available cores. This starting point assumes that only one application server is running on the operating system or logical partition. Setting the number of threads to twice the number of cores is based on experience that processing in Cúram is usually split relatively equally between I/O and CPU.

The way processing is then broken down between online and asynchronous processing depends on the characteristics of your system: how much asynchronous processing does it do? As a quick-start, a simple, equal split is suggested:

It is suggested that you set the minimum number of threads equal to the maximum number of threads. Setting the minimum threads equal to the maximum threads avoids the processing cost of pool growth and shrinkage. Use the following settings to configure the pools:

Then, monitor the thread pool usage and the number of threads that are adjusted according to CPU utilization. For example, if a thread pool is fully used and spare CPU capacity exists, you can add a thread. Spare CPU capacity, depending on the platform, might be CPU use below 80%. You need to define your CPU plateau threshold based on your environment configuration and the results of load testing.

In WebLogic Server (WLS) configure the two work managers that Cúram uses, which are the default work manager for HTTP requests and the MDBWorkManager for JMS. Specify a maximum thread constraint. As a starting point, set the maximum thread constraints to the following values:

Then, tune the work managers by monitoring thread usage. Monitoring indicates thread usage for a work manager. If all threads are used and CPU capacity exists, you can increase the maximum thread constraint.