[squid-users] Broken images and connection failures

I have been trying to troubleshoot a problem for over a week now, and
have resisted posting to the list in the vain hope I'll work it out
myself - however I am having no luck.

The problem is that during busy periods many pages are being displayed
with many broken images & frequently the browser (IE6) displays the
"Cannot find server or DNS error" message.

I am running Squid2.5STABLE6 on a Red Hat 9 (2.4.20-31.9smp kernel)
in an organisation which has approx 2500 users.
Average req/sec is about 80req/sec and peaks at 190req/sec during lunchtime.

The server is an IBM x345 with 2.5Gb RAM, 98Gb cache (only 16Gb used
so far), dual 2.4Ghz Xeon processors, with an Intel Pro 1000 connected
to a 1Gb switch.

I am doing content filtering with SquidGuard, NTLM authentication with
winbind from Samba 2.2.10 and wb_ntlmauth and external ACLs using
wb_group

I have googled and searched the list archives, but there is nothing
terribly recent that I can find.

I don't know if this is relevant, but /proc/net/sockstat shows up to
5000 TCP sockets in TIME_WAIT state in peak times.

Any suggestions to what may be causing this would be greatly appreciated.

Regards,
Rob Hadfield
Received on Wed Sep 15 2004 - 19:17:18 MDT