JVM Process Climbing Until Crash

Load balancer, to two proxy servers, to IS running on NT.

JVM processing climbs in percentage and does not decrease. Eventually, it hits 99%, the task manager hangs and have to reboot the machine. Once the machine is reboot there are about 40 sessions opened up (out of the blue, guessing it is from the proxy server). Things run well again for about a day until the JVM climbs and crashes the system again… Repo size is about 20 MB and there seems to be several unecessary sessions. Any ideas experts?

Any recent changes to your deployment?
Did you write any services that spawn a new thread?
Is there a chance you wrote a service that is executing a non-terminating loop (service A calls service B which calls service A)?

Also, do you have any remote servers configured and, if so, what timeout values are you using? I’ve seen infinite timeout settings (-1) on remote server definitions cause sessions to remain open.

Can you tell us the IS version and the JVM in use? If a newer JVM, have you tried the -server switch? Any mods to server.bat? Any scheduled tasks? Have you tried a scheduled task that runs garbage collection? (Mark C. will probably have a fit if you are–<g>) Is IS supporting users or just automated processes? Are the services being invoked over HTTP marked as stateless? If not, can they be? Does the server.log specify anything suspicious? Any meaningful entries in the error log? Using HTTPS? What are the session persistence settings on the load balancer? What is the session timeout on IS?

Just random thoughts.