bmurray at snf.stanford.edu
Tue Mar 6 15:33:39 PST 2001
John and Mike,
I have been intentionally running the Coral software for long
periods without restarting the servers because I wanted to stress test the
application and prove the robustness of our code. (Apparently Sun doesn't
have the same high standards.) Anyway, in most actual production
environments, Coral would be restarted once a week. So we probably shouldn't
waste any time tracking down this memory leak. We do have some options.
Until today, we had been running over two weeks under a heavy load with
no problems. First, we could leave the JVM initial memory allocation
for the event manager where it is and, additionally, set a mamximum
memory allocation to twice the intial allocation. That would probably
give us four weeks. That should be overkill.
More information about the coral