Re: Squid Performance Issues.

On Wed, 10 Dec 1997, Chris Keladis wrote:
> We have just recently installed about 20 Squid proxy servers running on
> either Solaris 2.5, or Linux 2.0.7..

That's a really weird kernel version to be using on Linux, from July 1996;
it may well have bugs that have long since been fixed, and 2.0.32 may well
perform better overall.

As for Solaris 2.5, it was highly unstable, don't use it, use 2.5.1.

> Now i am very unimpressed with speed of our Solaris/Linux squids, as
> compared to WindowsNT Proxy server v2.0 (bleah :-) )..
>
> The UNIX boxes are P100's, 32MB RAM, with 4 gig SCSI drives, using
> Adaptec AHA-2940 PCI Controllers, and 10mbit ISA Ethernet cards. I have
> timed the test loading identical web pages over a 26.4k modem connection.
> (Allways clearing the local browser cache).

A potential big variable is the speed of the serial connection *to* the
modem, which might conceivably vary from 9600 to 230,400 bps...you should
definitely try for 115,200 or better. Modem connections are such a tiny
bandwidth that there is not much room for proxying to play a role in
performance, other than an ICP parent (which you do not indicate you are
using) or caching effects (which you presumably controlled for?).

Also, ISA Ethernet cards suck, use at least a PCI NE2000, they are only
$20. What Ethernet cards are these?

> I have concluded Solaris was slightly faster than Linux, but on the WinNT
> machine (on the same hardware) it was almost twice as fast. I also tried
> FreeBSD on the machine and the performance was about the Same as Solaris
> (with identical squid.conf's).
...

If you got better performance under Solaris, that suggests to me you are
using both Solaris and Linux at 38400 bps between the OS and the modem,
because AFAIK that is all you can do easily on Solaris. You should use
115,200 or more to get the best out of the modem's built-in compression, I
suspect this is why NT is performing better.

I use Squid on Linux for web proxy purposes over a modem line as well, and
considering it gets very nearly theoretical maximum throughput I don't see
how NT could get double unless NT is using a 230,400 serial port on
incredibly compressible (null?) files! Also, going through the squid cache
from a Win95 client is noticeably faster than a direct modem connection,
which I attribute to having a nice fast upstream host the squid cache talks
to via ICP.

David
Niemi@tux.org 703-810-5538 Reston, Virginia, USA
    "Down that path lies madness. On the other hand, the road to
     hell is paved with melting snowballs." -- Larry Wall, 1992
Received on Tue Dec 09 1997 - 20:28:12 MST