OIM 9.0.1 Concurrent Usage Limitation - Performance
601723Oct 5 2007 — edited Oct 5 2007We're encountering an issue with concurrent users, specifically after about 25 concurrent users OIM becomes super slow to respond and in fact new users time out while trying to log in.
With the 20 to 30 concurrent users on the system, the JBOSS v4.0.2 thread count works its way up to about 350 of the allowed 500. OIM seems to have built up some kind of execution queue which it slowly chugs away at.
There appears to be a bottleneck somewhere in that the Java.exe CPU on that application server that OIM is running on is crusing at 2% utilization. In essence, OIM built up the queue but is not able to engage the CPUs to do the work faster.
About our sytem: Windows 2004 Advanced Server, 4 dual-core CPUs at 4Ghz, 4GB of RAM but JVM has 1.5GB allocated to it. The database is on SQL Server 2000 and is running on an adjacent similarly spec'd server but we've seen no bottlenecks there. The databse throttles at up to 30% CPU utilization when we have this type of load.
Performance Test - we got this 20 to 30 concurrent user figure from a performance test using LoadRunner. Loadrunner simulates fron-end users. It's worth noting that in one test LoadRunner crashed after it ramped up to the concurrent users which means that all the users dropped. However, OIM kept processing work for an additional 20 minutes after LoadRunner crashed. So this told us that there is a queue that OIM works on with 2% CPU Utilization therefore translating into the "slowness".
We've worked our way to this problem area where OIM cannot engage the CPU to execute its inner queue. We're looking into switch from JBossMQ to Jboss Messaging (http://wiki.jboss.org/wiki/Wiki.jsp?page=TuningJBossMQ) or the way we start OIM as a Service http://wiki.jboss.org/wiki/Wiki.jsp?page=RunJBossAsAServiceOnWindows .
We're currently using these to start the service: http://wrapper.tanukisoftware.org/doc/english/integrate-simple-win.html and http://forge.objectweb.org/projects/javaservice/
ANy thoughts on how that OIM inner queue can be controlled and to have it use more of the CPU horepower to process requests?