Re: [squid-users] Architecture

On Mon, 29 Jun 2009 18:05:22 -0300, Ronan Lucio <listas_at_tiper.com.br>
wrote:
> Adrian,
>
> Adrian Chadd escreveu:
>> Just another random datapoint - I've just deployed my Squid-2
>> derivative (which is at least as fast as Squid-2.HEAD) as a forward
>> proxy on some current generation hardware. It's peaking at 700
>> requests/sec and ~120mbit a sec with a ~ 30% byte hit rate.
>>
>> A reverse proxy with a high hit rate should do quite a bit better than
>> that.
>>
>
> Could you tell what hardware do you use?
> Reading Squid-Guide
> (http://www.deckle.co.za/squid-users-guide/Installing_Squid) it says
> Squid isn't CPU intensive, says a multiprocessor machines would not
> increase speed dramatically.
>
> I know this docs is so old, but it talks about machines like a Pentium
> 133 with 128 Mb RAM.

Followed by "while receiving 7 TCP requests a second".
If you have any sort of busy site, a single Squid can handle several
hundred TCP requests per second. This naturally uses a lot more CPU than 7.

FWIW when my squid 2.6GHz single-core box receives 2 TCP req/sec it also
uses very minimal CPU for the squid process ;)

>
> So initially I was thinking in Dual QuadCore + 4Gb RAM. Now I'm thinking
> in a Single QuadCore + 2Gb.
> What do you think about that?

The current Squid litany is a dual-core (one core for single Squid process,
one for OS + etc). Quad-core CPUs are currently wasted on a single Squid
instance. To make best use of them would be running 2-3 Squid instances.
Which implies a very high throughput requirement.

>
> I think a throughput like yours would be great for me.
>
> Another question: How many disks do you use?
> In other words: Do I need some special disk strategy to achieve such a
> throughput?
>
> Thanks,
> Ronan

Amos
Received on Tue Jun 30 2009 - 01:11:05 MDT