Re: [SQU] on multiprocessor machines

From: Robert Collins <[email protected]>
Date: Tue, 5 Dec 2000 20:13:11 +1100

With async io or diskd squid will use multiple processors much more effectively. IIRC Joe Cooper has some info on this..

Rob
----- Original Message -----
From: "Chemolli Francesco (USI)" <ChemolliF@GruppoCredit.it>
To: "'Rainer Ginsberg'" <rainer.ginsberg@de.bosch.com>; <squid-users@ircache.net>
Sent: Tuesday, December 05, 2000 7:32 PM
Subject: RE: [SQU] on multiprocessor machines

> > Hi all,
> >
> > so Squid doesn't scale very well on multiprocessor hardware. It's one
> > large process which can basically use only one CPU at a time.
> >
> > I was thinking about how to remedy this problem: My idea is to give
> > the machine as many IP addresses as it has processes. Then start the
> > same amount of Squid processes where each of them listens on one of
> > the IP addresses. Then you could apply load balancing, either via DNS
> > (cheap but not perfect) or with dedicated load-balancing appliances
> > (very expensive).
> >
> > The problem with this approach is that the Squid processes might not
> > "agree" on using different CPUs but all processes clogging to one or
> > two CPUs and thus reducing performance.
> >
> > Has anybody else thought about a solution for Squid on multiprocessor
> > systems?
>
> Thought - but not implemented:
>
> start as many squid processes as there are processors on different
> ports, then use some clustering solution (like Linux Virtual Server)
> to transparently redirect ports. All on the same system.
>
> --
> /kinkie
>
> --
> To unsubscribe, see http://www.squid-cache.org/mailing-lists.html
>
>

--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html
Received on Tue Dec 05 2000 - 02:07:51 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:56:51 MST