Re: [squid-users] couple of squid questions

On Tuesday 25 February 2003 09.40, Alex Sharaz wrote:
> hi all,
> couple of questions re squid config
>
> My squid hardware is based round systems with a 1.8Zenon cpu
> 4Gbytes of ram and 4 36Gbyte disks for cache storage running RH 8.0
> with the 2.4.18-24smp version of the kernel.

If possible with your motherboard you may want to consider using a
non-SMP kernel for increased system performance, but you probably
need to compile your own to support large quantities of memory.

There is a quite high performance penalty paid when enabling SMP in
the kernel, and the benefits of SMP is mainly seen if the
application(s) can really use more than one CPU. Squid is not at all
good at using more than one CPU.

> Each disk has 1
> partition set up cachedir ufs <90% of disk space> 256 256

You need more than 256 L1 directories for such large cache
directories..

   L1 = at least cache_size(in kb) / L2 / L2 / 8 * 2

> what sort of cache_mem setting should I be looking for in the above
> config.

Not too large. Even if your server have 4GB individual applications is
still limited by the 32bit pointer architecture of the Xeon processor
to about 1.5GB each in the best conditions..

See the Squid FAQ on memory usage for how to calculate how much memory
Squid will use for your cache directories. Then use no more than 50%
of what is left of the maximum process size for cache_mem.

> Should I be aufs?

Quite likely. If you get unexpected errors please report these errors,
exacly as they are given.

> given that the Zenon processors can support hyper-threading, is
> there any magical squid config settings that'll make use of this?

No. Squid is essentially a single threaded application.

Regards
Henrik
Received on Tue Feb 25 2003 - 02:46:45 MST