The VolanoMark 10/100 loopback test was ran to compare the 2.4.1 and 2.4.1+MQ Scheduler kernel. Previous Lockmeter data showed that the runqueue_lock had high utilization and spin time on an 8P system. The average run queue length on an 8P is about 70 and the processes run [on average] for a short period of time [there are a lot of sys_sched_yields]. The MQ scheduler patch was tested to determine if it could remove the runqueue_lock contention.
More information on the MQ scheduler can be found at http://lse.sourceforge.net/scheduling/
The 2.4.1 MQ scheduler patch improved throughput of the VolanoMark 10/100 loopback test.
8P system : message throughput increased by 150 % from 13,196 msg/s
to 33,078 msg/s.
4P system : message throughput increased by 37 % from 17,595 msg/s
to 24,213 msg/s.
Note that going from 4P to 8P no longer scales negatively.
Initial JVM heap sizes was 8MB.