Joseph Syu <[EMAIL PROTECTED]> writes: > We have noticed the memory usages on this gmetad server were mostly > consumed by apache services (each instance of apache process took > ~30MB memory), not by gmetad (only 6KB memory each) itself, so I > assume it is just the matter an under-powered machine. Am I on the > right track here? Or what may be the better way to architect a > system like this? please advice, thanks,
Ganglia is no big memory hog. Rather, it tends to be I/O-bound, since it has to update thousands of little bitty files all the time. What I did when our monitoring server had trouble keeping up was to put in a couple of fast disks (7200 rpm ultra-ATA 133), stripe them for performance and put a reiserfs file system on them to use as /var. This was in all certainty overkill, but it solved the problems; the RRD database files are once more updated correctly, and the web interface is, if not blazingly fast, at least usable. -- Leif Nixon Systems expert ------------------------------------------------------------ National Supercomputer Centre Linkoping University ------------------------------------------------------------
