Re: [squid-users] Re: Sun and Squid Cache

On Thu, 4 Feb 1999, Chris Tilbury wrote:

> On Thu, Feb 04, 1999 at 12:07:30PM -0000, Joerg Moellenkamp wrote:
>
> [snip]
> > Hmm, after nearly one and a half month with a ultra enterpise 250 under
> > solaris 2.6 , i think its a bad idea to locate squid on a sun ... at
> > least with slowlaris 2.6... when the cachedir utilisation is well under
> [snip]
>
> Before we enter into a vast debate about the pros-and-cons of Solaris
> for Squid, have you downloaded and installed Adrian Cockcroft's
> "SE Toolkit 3.0"? (http://www.sun.com/sun-on-net/performance).
>
> Cheers,

While I do agree that you need to choose space optimisation rather than
speed on UFS filesystems when running Squid 2 on Solaris 2.6, I shall
comment that I ran Squid 1 on the same hardware with exactly the same
disc setup with no disk problems for 18 months. Squid 2 seems to access
the disks differently in some way which tickles these UFS problems.

As for the performance problems? I bet that system is discbound. I'm
suspicious about the overhead that even hardware raid will be putting on
the disc subsystem. The SE Toolkit should show that.

I have two Ultra 1s, 170Mhz CPUs both doing a peak of 1500 TCP
requests/minute (which is approx half the load on that 250, but only half
the CPU).

Then again I don't try to run my disks to 100% capacity. For the extra
10% of cache you get a negligable increase in hitrate, and you make sure
that UFS gets stuffed. UFS requires some spare disk to operate
efficiently. My disks are 90% full with 1% minfree.

I'm also using the logging filesystem provided with ODS (and part of the
OS in 5.7) which gains me some of the benefits of the write cache on the
RAID system.

My median service time is under 400ms.

Suns may not be the best platform for Squid, but they're not as bad as
some people make out.

John

John Sloan
UK UNIX Systems Administrator
UUNET
An MCI Worldcom Company
Received on Thu Feb 04 1999 - 09:04:17 MST