Re: [squid-users] Recommend for hardware configurations

> On lör, 2008-07-05 at 12:44 -0300, Michel wrote:
>
>> I am not understanding why you keep suggesting single core as preferred cpu
>
> Did I? Not what I can tell.
>
> I said Squid uses only one core.
>

:) good answer ... but often it does not matter what we say but what is beeing
understood, what I meant is that it comes over as if you are suggesting single core
computers

>> even if squid's core is actually not multi-thread capable a faster cpu is better
>> -
>> there are also other things running on a machine so a smp machine ever is a
>> benefit
>> to overall performance
>
> Both yes and no. For an application like Squid you will find that nearly
> all OS:es gets bound to a single core running both networking and the
> application, leaving the other cores to run various tiny other stuff..
>

nope, not at all. probably on Linux's spin lock model it might be so , but I do not
know, on freebsd you can watch the squid process and it's threads, either aufs or
diskd related and see that they are handled by all cpus all the time

35867 squid 4 -19 3921M 3868M kqread 3 200:28 0.00% squid0
 1481 squid 4 -19 601M 581M kqread 0 86:03 0.00% squid1
 1482 squid 4 -19 598M 579M kqread 0 84:49 0.00% squid2
 1495 squid -4 -19 8300K 1376K msgrcv 1 20:19 0.00% diskd-daemon
 1496 squid -4 -19 8300K 1372K msgrcv 3 20:11 0.00% diskd-daemon
 1497 squid -4 -19 8300K 1324K msgrcv 3 5:42 0.00% diskd-daemon
 1498 squid -4 -19 8300K 1224K msgrcv 2 5:31 0.00% diskd-daemon

35867 squid 4 -19 3921M 3868M kqread 1 200:28 0.00% squid0
 1481 squid 4 -19 601M 581M kqread 1 86:03 0.00% squid1
 1482 squid 4 -19 598M 579M kqread 1 84:49 0.00% squid2
 1495 squid -4 -19 8300K 1376K msgrcv 0 20:19 0.00% diskd-daemon
 1496 squid -4 -19 8300K 1372K msgrcv 0 20:11 0.00% diskd-daemon
 1497 squid -4 -19 8300K 1324K msgrcv 2 5:42 0.00% diskd-daemon
 1498 squid -4 -19 8300K 1224K msgrcv 2 5:31 0.00% diskd-daemon

35867 squid 4 -19 3921M 3868M kqread 1 200:29 0.00% squid0
 1481 squid 4 -19 601M 581M kqread 2 86:03 0.00% squid1
 1482 squid 4 -19 598M 579M kqread 3 84:50 0.00% squid2
 1495 squid -4 -19 8300K 1376K msgrcv 1 20:19 0.00% diskd-daemon
 1496 squid -4 -19 8300K 1372K msgrcv 1 20:11 0.00% diskd-daemon
 1497 squid -4 -19 8300K 1324K msgrcv 2 5:42 0.00% diskd-daemon
 1498 squid -4 -19 8300K 1224K msgrcv 2 5:31 0.00% diskd-daemon

three tops in 3 different seconds, 8'th column show on which cpu it runs, observing
threads it still is more fun

>
> Why I recommend dual core instead of quad core is simply because you get
> a faster core speed in dual core than quad core for the same price (and
> often availability as well..) which will directly benefit Squid in high
> performance.
>

I understood you recommend single core ... not dual

> Yes, Squid quite easily gets CPU bound, and is then limited to the core
> speed of your CPU, and the faster the core speed is the better in that
> situation. Selecting a slower core speed to fit more cores hurts
> performance for Squid when the server is mainly for Squid.
>

I am not so sure if the core speed does matter so much as long as there IS CPU left
... then there is CPU left for any other work...

>
> You are welcome to give numbers proving that for Squid a 4 core system
> outperforms a 2 core system with the exact same configuration in all
> other aspects. Don't forget to include price in the matrix..
>
> The most interesting test configurations is
>
> - no disk cache
> - single drive for disk cache
> - 4 drives for disk cache
>
> Until I see any numbers indicating quad core gives a significant
> increase outperforming what the same price configuration using dual core
> I will continue propagating that quad core is not beneficial to Squid.
>

two/three years ago I said next year there are no single cores to buy anymore and
everyone is running at least dualcore if not quad and was shot by almost all
freebsd 4.x and dragon-fly lovers or should I say by people which didn't saw where
the modern threading model was going and were hanging on to the global giant lock
because at *THAT* time network and disk performance was still better?

then to be honest I do not believe that you ever will be convinced by any test *I*
post here :), so do it yourself and get your own conclusions ... the test is easy,
get yourself an AM2 MB and a X2 and a X4 and nuke a fixed rate of http requests
over a certain time into each CPU and monitor CPU time and disk io (on FREEBSD
amd64 7STABLE ) and compare it and I say you show me that X2 is losing and then I
get myself a linux box and shut my mouth :)

> Similarly for dual core vs single core, but it's not as clear cut as
> there is not a big per core performance difference between single and
> dual core compared to prices..

as I said soon there are no single cores to buy anymore, at least not for real
computers, so certainly price comparing does not count anymore and hardware is
cheap generally, actually there is no real price difference between an X2 and X4
and not only single versus X2

michel

****************************************************
Tecnologia Internet Matik http://info.matik.com.br
Sistemas Wireless para o Provedor Banda Larga
Hospedagem e Email personalizado - e claro, no Brasil.
****************************************************
Received on Sun Jul 06 2008 - 11:17:34 MDT