[squid-users] Another timeout problem..(gigahertz card)

I have a machine that was previously running RedHat 7.0 with
Squid 2.3.STABLE4-1, which I upgraded to 7.2 and Squid 2.4.STABLE1-6.

Because my boss wanted all our servers on a gigahertz backbone, I also
switched the network card from an Intel Etherexpress 100 to a Netgear
GA622T. In order to do that I had to upgrade the kernel, so I'm using
2.4.17.

Anyways, to the problem. The working configuration had no trouble at
all, but since making the switch, we get Read Timeouts all the time.
The odd thing is that the timeouts would occur instantly, so right
after you select a web page, the timeout page would appear right away.

The problem was so bad, that I tried downgrading Squid to the
2.3.STABLE4-1 version it was previously running. That helped the
problem quite a bit, but we still get it.

I've removed the cache_dir and had squid recreate it. I've tried using
the stock squid.conf from both versions of squid (the ones that ship
with RH7.0 and RH7.2). I've also tried the version on RedHat's erratta
page. Always starting from a clean slate by removing the cache_dir.
They all have the problem.

I've also tried not using the default timeout values by changing them to
what I think should be really high values, and the timeout page still
comes up instantly.

I noticed that the problem would seem to happen more frequently on
larger pages, so I tried increasing the values for the variables that
referenced size such as request_body_max_size. That didn't help either.

I've also double checked that I didn't compile in ecn by checking that
/sbin/sysctl -a |grep ecn had a value of zero.

I've even taken another RH7.2 machine, copied the kernel to it, and ran
squid 2.4.STABLE1-6 from RedHat's errata page on it. This other
machine has had no problem with any pages (granted I only moved ten
users on to it)

The only setting in the squid.conf file I changed from the default
was the cache_dir so that it would be on it's own partition.

The only thing I can think of is the fact that the server is using
a gigahertz card, that some autocalculated timeout is being hit because
the main pipeline to our business is not fast enough.

We have no other network problems internet wise. And the machine with
the gigahertz doesn't seem to have another network problem. It also
runs other services without a problem.

Any help would be greatly apreciated!

-Jim

-- 
James Lucha
Programmer/Analyst
San Bernardino Medical Group
E-Mail: ninerfan@sbmed.com
"Linux: Because rebooting is for adding new hardware"