Re: [squid-users] Squid work consideration

On 6/11/2011 12:50 a.m., Benjamin wrote:
> Hi Amos,
>
>
> While we are looking for cache_dir consideration , does a single big
> directory is better or multiple directory with good amount of size is
> better while we are looking for performance ?
>

Multiple physical disks with cache_dir on each is best. That offers the
most parallal reads/writes. Unless you are adding COSS or Rock storage
cache_dir its not worth placing multiple cache_dor on one HDD. Total
size of any given UFS cache_dir seems not to matter as much as
parallelism on the disk I/O.

> And i tried to find from internet that which is better for cache_dir
> file system ext3 / ext4 / reseizerfs ( in terms of heavy loaded
> systems ) ?
>

Disabling the journalling actions and atime updating is best for all of
the above. I have not seen any recent speed benchmarks comparing them.
If you base speed on the more generic benchmarks be aware Squid disk
access has a unusually high portion of writes.

>
> Whenever i m trying to find how many squid processes are running i got,
>
> pgrep squid
> 8311
> 8313
>
> it means squid always has 2 squid processes. Can we increase it for
> high performance.?

One of those will be the master process. One will be the actual worker
process.

You can run as many pairs of mster+worker as you like on a box. Though
we dont recommend running more than one worker per CPU.
See http://wiki.squid-cache.org/MultipleInstances for the relevant
config details for 3.1 and older. If you have squid-3.2 there is a link
there to the SMP support page.

Amos
Received on Sat Nov 05 2011 - 12:07:58 MDT