Is there a way to release the leaked memory without reboot the system?

[Daneil Goodman]

Hi There,

I have a Dell PoweEdge 1950 cluster running RHEL AS 4. The nodes are
internally connected using InfiniBand switch to let users to run the
parallel jobs. The cluster has serious memory leak issue which left a lot of
RAM and swap space behind after the job finished. I usually reboot the nodes
to clean up the garbage after I confirmed there is no job running on these
nodes. But recently I found sometimes other user's job could be killed when
I reboot the nodes although the killed job is running on other nodes. This
weird issue bothered me a lot. Therefore, I would like to find a way to
release the leaked memory and clean up the swap space without rebooting the
system. Can someone give me a suggestion?

Thanks a lot,
Goodman