Thank you for your reply to my post. I am replying to your e-mail and CC'ing the list so somebody else can make use of this thread...
I wish SOMEWHERE in the FAQ it would state that it is silly to run Squid on a multi-proc platform... What would the directives in my squid.conf look like in order to build it out like you described with 4 drives for cache and one week's worth of objects cached? I plan on re-building these two servers with two smaller servers but 2 x Pentium 600E Proc's and 1 gig of Memory each. Probably 5 x 9 gig drives. The reason for 2x Procs is that I believe that there are many other processes that can take advantage like the 20 NTLM web_group helpers I am running and the Apache server I am using for SARG and Webalizer reports. Squid runs decent with a second server added into the cluster load-balanced using a Foundry Server Iron but still not great. Also, I am running 4xIntel EE100Pro cards teamed together, not cheapo RealTech cards. No swapping so far, but I understand that the Squid memory footprint depends a lot on how the Cache Dirs are set up, so I want to avoid tapping swap at all costs. I am trying to take your advise to heart. Please advise. TIA. -Mark >1) You would have been MUCH better off with 4 individual systems. If it's > not too late, do that instead. Multi CPU is silly for squid. Actually, > you would be much better off with 2 servers. >2) Get RID of the RAID. You want 1 drive for boot and system stuff. The > logs can go on there if you want. Then I would go with 4 drives > for the cache. >3) Insure you are not swapping. Lots of people crank up their memory > utilization thinking that it will help a lot. It doesn't. You want about > 1 week cached. Don't be afraid to start with small cache and raise it over time. >4) Check your ethernet. Both switch and card should NOT be auto negotiate. > And you should be using decent Etherent cards (e.g. not RealTek) and > switches (e.g. not cheap Netgear). > >325 users should not be hammering you much at all. > >/\/\ \/\/ On Fri, Nov 21, 2003 at 01:15:40PM -0700, Mark Pelkoski wrote: >> List, >> I FINALLY implemented a Squid server into my production environment >> today. It is squid-2.5.STABLE4-20031029 on a Quad Proc Xeon 500MHz >> with 1M Cache and 3x9Gig Raid-5 dedicated for cache and 2x9Gig Raid-1 >> for OS Redhat 9.0. I tested this exact server config with 74 users and >> performance was pretty good for 800 Reqs/Min. Now in Production I have >> 325+ users at 2700 Reqs/Min and performance stinks. It's like being on >> 325+ a >> dial-up connection. 3 of the procs are sitting below 1%. The other is >> used by Squid at 99.9%. It there any way to speed up performance on a >> multi-proc system? TIA.
