i'm in a similar situation. i'm replacing an isa box (something like a 2ghz machine with 1 big ata disk) with squid over the summer.
we have a dual p2-400 server with 3 scsi disks in. i'm in the process of putting squid on it now using a 686 smp kernel to try and squeeze every ounce of speed from it!
one thing i can tell you at this stage, is that if you set up software raid0 to take care of the striping (rather than configuring squid to use the different drives) the access is way ahead of the performance of a single ata drive. and this is from a server which is around 6 years old.
incidentally, we use a separate box for dansguardian with a fast processor in it. you really need that for dg.
john
--- On Fri 06/17, Chris Robertson < crobertson@xxxxxxx > wrote:
From: Chris Robertson [mailto: crobertson@xxxxxxx]
To: pieterjan.heyse@xxxxxxxxxxxxxxxxxxxxx
Cc: squid-users@xxxxxxxxxxxxxxx
Date: Fri, 17 Jun 2005 10:53:00 -0800
Subject: RE: Performance question
> -----Original Message-----<br>> From: Pieterjan Heyse [mailto:pieterjan.heyse@xxxxxxxxxxxxxxxxxxxxx]<br>> Sent: Friday, June 17, 2005 4:01 AM<br>> To: squid-users@xxxxxxxxxxxxxxx<br>> Subject: Performance question<br>> <br>> <br>> Hi folks,<br>> <br>> we are having some kind of performance problem here and I want to ask<br>> you guys for advice.<br>> <br>> We are a school with 1200 students, and 200 pc's running squid +<br>> Dansguardian as proxying service. The http requests tend to be slow at<br>> peak moments and I am wondering if I could resolv this by changing the<br>> setup.<br>> <br>> Currently we have a celeron 900 pc with 1 ATA disk acting as a proxy.<br>> A nearby factory donated an old dual P2-300 server with 7 SCSI disks<br>> 4GB each. I read that performance of squid increases with every disk<br>> you add, so will the switch to this slower (cpu wise) system be<br>> beneficial to us, ro should I stick to the celeron 900 and add some<br>> more RAM
?<br><br>Perhaps you should use both. Put your cache on the new (slower) server, and<br>run DG on the old (faster) one. Squid by itself does not use much CPU, and<br>is primarily limited by IO. On the other hand, since Squid does not take<br>advantage of multiple CPUs DG would have its own processor on the new box.<br>I haven't run it, so I don't know what it requires from the CPU. 200<br>clients does not sound like much of a load for one P2-300 server.<br><br>> <br>> The celeron has 256MB RAM, the donated server has 512MB RAM.<br>> <br>> Are there ways to benchmark squid in an easy way ?<br><br>Not that I know of. Polygraph, while very informative, is not (in my<br>opinion) easy. YMMV.<br><br>In any case, running top and vmstat (again, assuming *nix), and looking at<br>the output of the cachemgr.cgi should go a ways towards showing you where<br>the bottle-neck is.<br><br>> <br>> Thanks,<br>> <br>> Pieterjan Heyse<br>> <br>> <br>> ICT Coördintor KSGWL -
Scheppersinstituut <br>> Scheppersinstituut Wetteren<br>> Cooppallaan 128<br>> 9230 Wetteren <br>> Tel: 09 3692072<br>> Fax: 09 3661348<br>> mailto:pieterjan.heyse@xxxxxxxxxxxxxxxxxxxxx<br>> <br>> <br><br>Chris<br>
_______________________________________________
Join Excite! - http://www.excite.com
The most personalized portal on the Web!
