SlavaNAP has a known memory leak introduced by the Borland C++ compiler. It doesn't always manifest itself, but I know my node, which handles about a 1000 users is only good for two weeks or so before a reboot is needed. Killing the process doesn't help much, as the OS doesn't seem to reclaim ALL memory, so slowly there is less and less available, and everything gets slower over time.
Hmm, I doubt this is a problem with a 50 user server however.
|