On 31.05.2012 01:26, Christoph Fink wrote:
Hi Squid Users,
I have a server running Debian Squeeze and Squid3 (version 3.1.6 from
Debian repo). It is configured using winbind+ntlm for authentication.
The server ran fine for a while now (I'd say 6 months more or less)
but
now I get a lot of "TCP: time wait bucket table overflow" errors in
syslog. After ~30 minutes the system then begins killing processes
and
reboots. First I thought that the problem was that the server had not
enough RAM. I upgraded from 512MB to 1GB but the problem still
existed.
Something unspecified is still consuming all the RAM you gave the box.
Might be Squid, 3.1.6 has several memory leaks fixed in the latest
releases. Try installing squid 3.1.19 package from the Wheezy/Testing
repositories. It has a lot of NTLM connection fixes as well as better
memory management.
Then I controlled on the VMware Host (the server is a virtual machine
running on a ESX Cluster) about the performance datas and I saw that
performance was not the issue (Memory, CPU and all the other counters
are at 30% max.). About 50 people are using that proxy Server.
Number of end-users is irrelevant to HTTP. Number of requests per
second is the metric which counts.
How many requests per minute/second is Squid cache manager reporting?
There are no other services running on that Server. It's just squid3,
samba (for winbind) and apache (only hosting the proxy.pac).
Squid Configuration file:
http://pastebin.com/5yngPt3c
What is this "use_negotiate_ntlm" configuration? Not something squid
supports.
Amos
