thanks for the advice Amos but I've just reset all objects in my cache (thus it's empty now) and the CPU utilization has decreased dramatically (up to 17%), but clients still get (rarely) error that the proxy is refusing connection. 2009/03/22 13:34:42| httpAccept: FD 8: accept failure: (53) Software caused connection abort 2009/03/22 15:16:49| httpAccept: FD 8: accept failure: (53) Software caused connection abort I'll try your advice in my testlab first before go into live server. thanks in advance. --- On Fri, 3/20/09, Amos Jeffries <squid3@xxxxxxxxxxxxx> wrote: > From: Amos Jeffries <squid3@xxxxxxxxxxxxx> > Subject: Re: access to cache is very high > To: "nyoman karna" <balique8061@xxxxxxxxx> > Cc: squid-users@xxxxxxxxxxxxxxx > Date: Friday, March 20, 2009, 10:19 PM > nyoman karna wrote: > > guys, > > > > I've been using squid for 5 years in my educational > institution > > (IM Telkom - Telkom Institute of Management) > > but this is the first time i encounter this problem. > > need your expertise. > > > > I'm using HP Proliant ML110 G5 with 320GB of SATA > disk. > > > > I'm using squid 2.6 stable 16 (FreeBSD 6.3) with 100GB > of squid-cache. > > when the cache was not yet full, there's no problem > whatsoever > > but when the cache is full (but still 90% of the > mountpoint), > > squid is always busy validating its objects > > and refusing all client connection. > > > > this is part of file /usr/local/etc/squid/squid.conf > > ------------------------------------------------ > > maximum_object_size 16384 KB > > minimum_object_size 0 KB > > maximum_object_size_in_memory 8 KB > > cache_replacement_policy lru > > memory_replacement_policy lru > > cache_dir ufs /webcache 100000 64 256 > > > > this is the capture when the squid refusing > connection > > -------------- (using top) > ---------------------------------- > > CPU states: 39.5% user, 0.0% nice, 55.3% system, 5.3% > interrupt, 0.0% idle > > PID USERNAME THR PRI NICE SIZE > RES STATE TIME WCPU COMMAND > > 77059 squid 1 123 > 0 186M 184M RUN 0:10 86.26% squid > > > > -------------- (using systat) > ---------------------------------- > > /0% > /10 /20 /30 /40 > /50 /60 /70 > > ad4 MB/sXXXXX > > > tps|XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX > > > > -------------- (cache.log) > ---------------------------------- > > 2009/03/20 09:54:44| httpAccept: FD 11: accept > failure: (53) Software caused connection abort > > > > > > UFS as the slowest IO process in Squid, is not the best for > large caches. Since you are on FreeBSD try diskd instead. No > change of cache is needed to alter ufs->diskd as the > storage manager type. > > Average HTTP object size these days is between 64KB and 128 > KB. Your max in-memory size of 8KB is causing a lot of > objects to be disk-saved without need. > > Also check the garbage collection range you have set: > http://www.squid-cache.org/Doc/config/cache_swap_low/ > http://www.squid-cache.org/Doc/config/cache_swap_high/ > > > Amos > -- Please be using > Current Stable Squid 2.7.STABLE6 or 3.0.STABLE13 > Current Beta Squid 3.1.0.6 >
