Thanks for all the data. One thing I'd like to suggest is getting the Java Thread Dump (send a
kill -3 to JVM process) in order to get the same (it goes to domain/logs/jvm.log), when you
experience the CPU spikes.
The number of Grizzly threads seem to be slightly higher than usual. With 8 cores, your max
number of Grizzly threads go to 250, which is rather high. That's just an observation.
It looks to me that you can consider sending "-d64" instead of "-server".
Also, what is the average time per HTTP request that is spent in the database tier?
Some other data like pfiles would help too. But a number of stack dumps are required at a minimum.
Regards,
Kedar
[Message sent by forum member 'km' (km)]
http://forums.java.net/jive/thread.jspa?messageID=338601