[salims]

We are load testing OTMv5.0 on Red Hat linux 4.0AS envionrment with JRocket JVM 1.4.2. Application server is running HP BL680c with 16 cores.

We are seeing good performance results but finding planning thread group not scalaing under load. It seems to be not using the cores fully at all.

Our observations are

The AgentUtility thread group is processing transactions 4-5 times faster under load, however, the AgentPlanning is 4-5 times slower, taking an average of 10 seconds to process transactions when set to 40 threads, and about 5 seconds per transaction with 20 threads. Further to that, I reduced the threads to 10, and again saw the time half.

Again I reduced the threads to 5, and saw the time reduced to 0.24 seconds per transaction.

It appears that the AgentPlanning Thread Group is only able to utilise one core on the Application server(bl680c02). All other thread Groups are scaling.

Following are the JVM parameters from weblogic.conf
wrapper.java.additional.1=-jrockit
wrapper.java.additional.2=-Xms2304m
wrapper.java.additional.3=-Xmx2304m
wrapper.java.additional.4=-Xgcprio:throughput
wrapper.java.additional.5=-Xverbose:memory
wrapper.java.additional.6=-Xnoclassgc

Are there any other tunning required on Linux for it to scale better?

[chrisplough]

Salim,

Could you post a follow-up, based on the latest round of performance testing? I know that we found a single-threaded issue related to the way BNG was being utilized, but believe the community would benefit from your experience testing and comparing the performance of Linux versus Solaris.

Thanks,
Chris