Re: [squid-users] Squid on DualxQuad Core 8GB Rams - Optimization - Performance - Large Scale - IP Spoofing

From: Tek Bahadur Limbu <[email protected]>
Date: Wed, 17 Oct 2007 14:46:18 +0545

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi All,

I would also like to share my thoughts regarding the hardware specs for a Squid proxy server which I am posting below.

On Wed, 17 Oct 2007 08:03:22 +0100
Tony Dodd <tony@last.fm> wrote:

> Adrian Chadd wrote:
> > Out of curiousity, how many Squid servers do you have deployed out there?
> >
> >
> >
> > Adrian
> >
>
> We've got 8 in total, currently. I'll preempt you asking for specs:
>
> 4 are:
>
> Supermicro 1u with Dual Core Xeon 5148 2.33Ghz, 4gb DDR2, 4 x 400gb
> 7200rpm disks in hardware raid 1+0.
>
> These guys proved to be too slow; they start hitting I/O overloads at
> around 100-150 requests/sec, so I got:
>
> Poweredge 1950s with one Quad Core Xeon L5310 1.6Ghz, 8gb FB-DIMM, 4 x
> 73gb 15krpm SAS drives in hardware raid 1+0.
>
> I haven't actually been able to hit the performance limits of these
> machines yet; I capped out at a kernel limit around 400 requests/sec.
> Interestingly, these guys only cost $200 more than the poor spec SM
> machines.

Well I think that it's not only hardware specs that we have to consider. We also have to take into account the operating systems, optimizations, Squid versions, Squid's conf files, gateway routers, etc...

One of my hardware is a refurbished Dell OptiPlex GX-270 purchased at around $200. Technically this is not a server but rather a desktop!

It has the following specs:

<DELL GX270 >
OS: FreeBSD-6.2 (i386)

38146MB 7200 RPM IDE hard drive
38146MB <Seagate ST340014A 3.16> at ata0-master UDMA100
CPU: Intel(R) Pentium(R) 4 CPU 2.80GHz (2793.01-MHz 686-class CPU)
real memory = 1072103424 (1022 MB)

/dev/ad0s1d 10154158 6294544 3047282 67% /cache1

With this hardware, my proxy server can easily serve 60-80 req/sec (3600 - 4800 req/min). If I push it, it can serve upto 150 req/sec (9000 req/min).

The median response service time hardly cross 1.3 seconds considering that we have a satellite link.

The CPU utilization which is always less than 15% suggests that it can serve more requests than what it is currently serving.

Attached are the current Request/Service/CPU RRD graphs for this proxy server.

But I don't mean that we should disregard good and expensive hardware but not everybody can afford them due to some restrictions and constraints.

I would love to have a IBM System P series server someday!!

>
> The squid cluster is set up to only talk to origin servers, and they
> don't have a sibling relationship; I found that the 1-2 second overhead
> for query/fetch from siblings was impeeding performance... and screwing
> up my graphs with leaps to 2000msec from the usual 10msec response time.
> In front of the squids, we have lvs + perlbal, depending on the domain
> being accessed. I should also mention that they're in use as a reverse
> caching proxy.

Are your siblings on different subnets?

Thanking you...

>
>
>
> --
> Tony Dodd, Systems Administrator
>
> Last.fm | http://www.last.fm
> Karen House 1-11 Baches Street
> London N1 6DL
>
> check out my music taste at:
> http://www.last.fm/user/hawkeviper
>

- --

With best regards and good wishes,

Yours sincerely,

Tek Bahadur Limbu

System Administrator

(TAG/TDG Group)
Jwl Systems Department

Worldlink Communications Pvt. Ltd.

Jawalakhel, Nepal

http://wlink.com.np

http://teklimbu.wordpress.com
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (FreeBSD)

iD8DBQFHFc9bfpE0pz+xqQQRAhfuAJ0cvLNLib5w0TBTYbnYcgKvw38BRACeKM6n
47mme7N7E55O36Zh6nbDA1c=
=2l5a
-----END PGP SIGNATURE-----

cache7.connections.day-17-oct-2007.png cache7.cpu.day-17-oct-2007.png cache7.svctime.day-17-oct-2007.png
Received on Wed Oct 17 2007 - 03:01:30 MDT

This archive was generated by hypermail pre-2.1.9 : Thu Nov 01 2007 - 13:00:01 MDT