Kelsey Cummings
kgc@corp.sonic.net
Thu Oct 26 16:46:14 UTC 2006
On Wed, Oct 25, 2006 at 09:20:29PM -0600, Vernon Schryver wrote: > > From: Kelsey Cummings > > > I run 4 dcc servers and have run into something confusing. On 2 out of the > > four servers that run dccd the RSS of the dccd process is 1.6g with a > > vmsize of ~2g. On the other 2, the RSS is only a couple hundred mb where > > the vmsize is also ~2g. It appears that on these two servers dccd is also > > trashing the disks pretty hard whereas on the other two, little or no disk > > activity is shown. The only difference that I've been able to figure out > > so far is that the two servers that with the lower RSS are public servers. > > The other two are not but are configured identically. > > > > Any ideas here? > > Since the 4 boxes are similar, and the 4 dccd process have similar > virtual sizes, the reason that their resident sets differ can only be > (a) one pair of servers is hardly working and so their residents are > small or (b) other work is fighting for RAM and squeezing the pages out > memory. Lots of disk activity is consistent with thrashing, so > my guess is (b). I find `top` for seeing what's going on. > And sometimes `vmstat 1` Or the local UNIX flavor's equivalents. The only problem with that Vernon is that the two servers which are hardly working have the larger resident size. The public servers only a 170-250mb. Of course, this morning, all of the servers are using about 1.5gb -- yesterday the unusual memory usage was persistent even after restarts of dccd. I'll keep an eye on it. -- Kelsey Cummings - kgc@corp.sonic.net sonic.net, inc. System Architect 2260 Apollo Way 707.522.1000 Santa Rosa, CA 95407
More information about the DCC
mailing list