Seeing significant performance degradation when using dynamic http thread pool when running a JSP HTTP micro benchmark. The degradation has been traced to a hot ReentrantLock in SyncThreadPool implementation of thread pool. When FixedThreadPool is used, there is significant improvement in performance results compared to dynamic thread pool.
Have tried a patch from Alexey replacing ReentrantLock with a Synchronized block, but that is not yielding much improvement.
The benchmark is a simple war file with 3 JSPs doing HelloWorld, showing current date and maintaining a simple user session with hit count.