This section shows configuration properties that you can change to handle the application which allocate 5,000 to 30,000 partitions. In general, you should keep the number of actual partitions to a minimum to ensure the cluster performance is not degraded. In addition, each attribute column has a footnote explaining how the attribute can be set.
For a maximum configuration IBM has tested, 10,000 partitions spread across 30 application servers. Following are the recommended configuration changes.
Update core group coordinator settings using the wpfadmin parameter, updateCoreGroupCoordinators. Set the -numCoordinators option to 4, and set the -preferredCoordinatorServers to be on different physical nodes as described earlier.
Update the highly available (HA) manager service transport buffer setting for all servers in the cell using the wpfadmin parameter, updateHamConfig. Set the -buffer option to 20 and set the -cell to the name of the cell.
Change the maximum heap size for all servers in the cell to 512 and set the deployment manager's maximum heap size to as large as possible (1400 if you have the physical memory). This can all be done through the WebSphere GUI and requires restarting all modified servers.
If you see OutOfMemory, JMXTimedOut, and other errors when running wpfadmin commands, try increasing the number of coordinators for the core group. This may be seen with large numbers of partitions.
Partition Count | Number of Coordinators |
---|---|
1000 | 2 |
5,000 | 2 |
10,000 | 3 Note: Use the HA manager policy attribute to collocate
coordinators exclusively on application servers located on different physical
nodes (see wpfadmin).
|