Recently, my application server (websphere) running under RHEL 3 starting to hang (able to connect but won't provide any response) under heavy load. We observed (using top) that during that time all memory (my server have 4GB RAM) have go to cached (around 3GB) and the available RAM drops below 600M and sometimes even worst that only 60M is available.
To resume the service, we have tried to kill the websphere process using kill -9 pid and then try to start it again. While we observe that after killing the websphere process (actually is a bunch java process). The available memory didn't change at all and this makes starting the websphere server failed.
Any way that I can control the use of cache memory ?