Re: [squid-users] Squid with dual processor ..

On Mon, 7 Jul 2003, Wei Keong wrote:

> Hi All,
>
> Good day. Just wondering if anyone has tried running Squid 2.4/2.5 on
> hyperthreaded Intel P4 CPU? Could this be the interim solution?
>
> Rgds,
> Wei Keong

Hyperthreading relates to making multiple virtual CPUs available to
processes, so the same inherent threading issues will exist as with
running on non-hyperthreaded systems.

In fact, there could be a performance hit if, as in FreeBSD (pre 5.3), the
scheduler is not "intelligent", and allocates processes in a round robin
fashion: the best possible scenario in a SMP system is if the squid
process gets its own CPU, and the other(s) get most of the other system
tasks; in hyperthreading, CPU allocation is overloaded, so anonymous tasks
could get allocated to a CPU that is also running Squid, causing Squid to
get swapped out.

Of course, when scheduling allocation and controls improve on such
systems, there may be benefits, but this appears to be some way off at
present.

J.

                     _______
+------------------ //////// -----+----------------------------+
| John V Cougar __ | Voice: +61 2 6208 1683 |
| Cache Manager / / /\ +----------------------------+
| Telstra Internet \/_/ \ Direct | E-Mail: cougar@telstra.net |
+--------------------____/ -------+----------------------------+

        
>
> On Sun, 6 Jul 2003, Henrik Nordstrom wrote:
>
> > On Sunday 06 July 2003 19.53, Schelstraete Bart wrote:
> > > That's maybe a good suggestion for Squid v3.0..........
> > > multitasking = work with multiple CPU's
> > > That should be something nice, as Squid is almost always used on
> > > servers.......and most server have multiple CPU's.
> >
> > Squid-3.0 will not improve in this aspect. The hope is that Squid-3.0
> > will be somewhat faster than Squid-2.5, but it sill will use a single
> > CPU thread only.
> >
> > There is much thought going on in how to make Squid SMP scaleable, but
> > so far all ideas requires rather far going changes to the internals
> > of Squid, more than what is practically possible in one development
> > cycle.
> >
> > The Squid internal design is gradually being refined and cleaned up
> > however, and it is not unlikely there at some time will be a SMP
> > capable Squid available, it won't be Squid-3.0 however, and quite
> > likely not Squid-3.1 either (at least not fully).
> >
> > The Squid-3.0 release is selected as a middleground, with a whole lot
> > internal redesign most users won't notice but preparing for the
> > development of more visible news in later releases, combined with a
> > reasonable amount of noticeable new features much asked for.
> >
> > Notice: Very little of the founded work in Squid-3.0 is performance
> > related. Allmost all is functionality related. The performance
> > related work and research has been done almost entirely by individual
> > developers own interest alone.
> >
> > Regards
> > Henrik
> >
>
Received on Sun Jul 06 2003 - 22:14:13 MDT