I have rerun my test and captured several thread dumps during the run.
From what I can tell, comparing the thread IDs to the output from prstat, the load is occurring randomly across my worker threads -- I believe as it should. However for the load applied to my SUT, it is seemingly with excessive and climbing cpu usage.
A question:
My load is pointed only at http on port 8080. with the usual listeners on 8181 and 4848. Why do I see heavy loads on the threads for the httpSSLWorkerThread-8181 and httpWorkerThread-4848?
snoop shows no activity on ports 8181 and 4848.
By the way, after following the earlier tip from h1d3m3, I found that the TDA plugin in VisualVM (using remote jmxrmi connection) directly shows the thread ID matching the ID shown in prstat. This avoids having to run pstack which is quite intrusive on the test run.
[Message sent by forum member 'russpet' (russpet)]
http://forums.java.net/jive/thread.jspa?messageID=318194