i'm monitoring median_select_fds
this morning with 150req/sec
select_loops = 280.262863/sec select_fds = 1502.051748/sec average_select_fd_period = 0.000660/fd median_select_fds = 3.984375
thin evening with 40req/sec
select_loops = 383.217992/sec select_fds = 457.205789/sec average_select_fd_period = 0.001830/fd median_select_fds = 0.000000
Duane Wessels ha scritto:
and actually configuration:
cache_mem 64 MB cache_swap_low 85 cache_swap_high 90 maximum_object_size 65536 KB maximum_object_size_in_memory 24 KB cache_replacement_policy heap LFUDA memory_replacement_policy heap GDSF cache_dir diskd /var/cache/spool/0 28000 96 256 Q1=72 Q2=64 cache_dir diskd /var/cache/spool/1 28000 96 256 Q1=72 Q2=64 cache_dir diskd /var/cache/spool/2 28000 96 256 Q1=72 Q2=64 memory_pools_limit 50 MB cache_access_log /var/cache/log/access.log cache_log /var/cache/log/cache.log buffered_logs on
Squid sames to work fine, but cpu usage after ~60Req/Sec on each server
is always 100%
I have tried many changes to squid.conf with no results.
Probably the first thing to look at is whether or not the high CPU problem comes from select/poll looping very quickly. For example:
% squidclient mgr:5min | grep -i select select_loops = 72.180305/sec select_fds = 19.528906/sec average_select_fd_period = 0.007477/fd median_select_fds = 0.000000
I think "median_select_fds" is essentially broken and always returns 0.
I plot the values for my caches, which you can see here: http://www.ircache.net/Statistics/Vitals/rrd/cgi/select.day.cgi
You might want to add 'half_closed_clients off' to your config file and see if that helps.
Duane W.
--
*Giulio Cervera*
EDS PA SpA Via Atanasio Soldati 80 00155 Roma (Italy) tel: +39 06 22739 270 fax: +39 06 22739 233 e-mail: [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>