Re: [squid-users] tweaking squid values

Roger 'Rocky' Vetterberg wrote:
> Hi list.
>
> I've recently been given two spare servers that are to be deployed
> as squid caches for a network with about 300 workstations, all heavy
> internet users and sharing a 10MBit dedicated internet line.
> The idea is to have the squids proxy http and MSN messenger
> connections, as well as some ftp restricted to certain sites.
>
> The servers in question are one PIII 1.0GHz with 1G of RAM and 72G
> of raid-0 diskspace, and one PIII 1.4GHz with 512M RAM and 36G of
> raid-0 diskspace. Both run the latest 6.x version of FreeBSD and
> Squid 2.5.14_2.
>
> I have tried to configure them according to what I read in
> documentation and FAQ's, but I run into heavy swapping, "Unable to
> allocate" errors or just bad performance. It seems I'm having
> problems finding a good balance between performance and stability.
>
>
Have you read through the genuine Squid FAQ section on memory
(http://wiki.squid-cache.org/SquidFaq/SquidMemory)? It's probably a
good place to start.

> Could someone give me some rough figures to use for cache_mem,
> cache_dir, L1, L2, Q1 and Q2?
>

Personally, I leave cache_mem at the default value, and trust my OS to
cache disk accesses. As for the cache_dir, the consensus seems to be
not to fill your partition beyond 60% for best performance.

> Would I benefit from using diskd, or should I run with normal UFS?
>

Use diskd or aufs.

> Both servers are armed with hardware raid on 15k drives, so the disk
> I/O should be pretty decent.
>

Just as long as you are aware that RAID 0 really doesn't have the "R",
and as far as I am aware, most RAID controllers only show significant
performance increase when calculating parity (RAID 3-6), you might be
better off using one spindle for the OS and creating separate cache_dirs
on each of the other spindles. That way a SPECIFIC disk has to die for
your proxy server to go down.

> TIA
> --
> R
>
>
Chris
Received on Wed Aug 23 2006 - 13:28:46 MDT