On Tue, 2006-02-07 at 15:36 -0800, Gregori Parker wrote: > I just deployed 3 squid servers in a similar configuration > (reverse-proxy serving large media files)...except each server of ours > is dual 3Ghz Xeon, 64-bit everything, 4GB RAM and around a TB each of > dedicated cache space (aufs on ext2 with noatime option). They are > running Squid 2.5 STABLE12 on Fedora Core 4 x86_64. Disk performance > looks fine to me, but I'm concerned because top reports that squid is > averaging 70% cpu usage most of the time. That percentage also includes time spent in the kernel? Check your overall system CPU usage stats for that. > Can anyone recommend techniques for assessing squid performance? I > have no good way of benchmarking our clusters since SNMP isnt ready > quite yet. Please don't mention cache_mgr, thanks :) top(1) is your friend. Compare the squid CPU usage with the time spent within the kernel. In some very busy systems I have built, Squid used up to 95% CPU, but half that time was actually spent within the kernel doing network i/o, poll(3) table scanning etc. Kinkie
