Re: Help me urgently : How can I speed up our squid?

From: John Lauro <>
Date: Tue, 03 Feb 98 08:33:16 -0500

On Tue, 3 Feb 1998 14:37:37 KST, 이종경 wrote:

> Thank you very much for your concerns!
>Now, we have the 2GB physical memory per Squid server and the 64GB
>Physical Disk(Hitachi Disk Array-DF300), so I allocated cache_mem
>as 896MB and was using the 15 cache_dirs on Disk array.
>But, All Squid Servers(EP3000*3) are very slow although doing well
>load balancing of the squid servers(DNS Round Robin) and Disk IO per
>squid server. Some guys recommended to reduce the cache_mem, but I don't
>know how much allocate for squid cache_mem.

Everything looks good. You do have over a magnitude of requests/hour
over me, and I don't recall seeing anyone state they were getting that
much traffic... The only thing you haven't shown is how the stats of
'iostat -cD -l 99 1' are during peek time. (Run for at least 30
seconds). That might help show where the bottle neck is.
Unfortunately it will not say how to best fix it, except by saying if
it's the CPU or disk I/O overloaded, or a disk (such as swap, or where
logs are) out of ballance.

You can dropping off all the children, siblings, and parrents to see if
that will help drop the load enough to keep from stalling. Dropping
cache_mem to 250 *might* help give solaris more disk buffers to use and
let it better deal with your heavy disk I/O. However, with 2GB of RAM,
I doubt it will make more then a few % difference....

Maybe you reached the limits of the current version of squid, and need
to consider the commercial software options.

Novell's bordermanager claims over 4,000 connections /sec. (See However,
that would be totally different hardware, etc... (I doubt if NT could
come close to that for their proxy server....)

NetCache from netapp is supposed to automatically turn off writes if
the drives cann't keep up (read only mode), and do other things to keep
from slowing down during very heavy traffic. They have a free trial
for 30(?) days, and may be worth trying. See: With the
exception of lossing your current cache, and redoing a config file it
should drop in nicely on your existing hardware for testing.
Received on Tue Feb 03 1998 - 04:58:38 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:38:46 MST