Re: [Fwd: Optimizing Squid performance]

From: Duane Wessels <[email protected]>
Date: Mon, 7 Aug 2000 16:55:48 -0600 (MDT)

On Mon, 7 Aug 2000, Athanasios Kouroussis wrote:

> Hi,
>
> I am running Squid 2.4devel-ntlm3 under Linux 2.2.14 on a Compaq
> Proliant 1600 (1 PIII 600Mhz, 512Mb RAM, 4x9GB SCSI-3 hard disks on a
> single RAID 5 logical unit controlled by a SMART-2 RAID controller).
> I am attending approx. 1000 users and I am getting (through cachemgr)
> about 2000hits/min at peak time. I am using squidGuard a redirector for
> filtering banned sites and a home made authenticator for NTLM users.
> CPU utilization is low (normally above 90% idle state).
>
> 1) The whole 512Mb of RAM are in use most of the time (even at 3am when
> only 10-20 users are active). Is this normal? The box used to have 256Mb
> (also all in use) and I notice no difference in performance after
> duplicating the size of the box's RAM.

Its hard to say if its normal. How did you measure memory usage?

> 2) I would wish to fine tune my box for maximum performance. Although it
> can attend 2000hits/min I think it can do better than that. Is there a
> certain manner of knowing that squid is the bottleneck and not my 2Mb
> internet link? What should I change in the conf file and/or OS and /or
> box in order to achive max. performance? The conf file is almost
> untouched except the dir locations and acl sections.

First, never use RAID. Use boring unix filesystems, one per
disk drive.

> 3) I use 10 authenticators. If I shut squid down, it does not kill the
> authentication chldren. Instead they remain alive, with PPID 1 and
> consuming all of the box's CPU (each 10% aprox). Do I have to take any
> special considerations in my authenticator in order for squid to kill
> gracefully the processes when going down?

Sounds like your auth process does not exit when it receives
an end-of-file condition.

Duane W.
Received on Mon Aug 07 2000 - 16:58:39 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:54:45 MST