On Sat, 3 Apr 2004, Steve wrote:
> The problem is that we have very poor performance, and have a group of
> people taking the shotgun approach to its cause. I've ruled out aliens,
> sun-spots and nostradamus, but I can't refute claims of corrupted cache,
> index issues, etc, because I don't know enough about squid. I have been
> told that we need to reboot the server down every day, delete the acct file
> every week, and start squid with the "-z" option every 3-6 months. I don't
> know if this is SOP, or if I'm supposed to do it because its the way things
> have always been done.
Probably just bad habits from not trying to deal with a slight
misconfiguration of the server.
The symptoms you describe is quite commonly seen if your server runs short
on memory, usually due to Squid being configured too optimistically (see
FAQ on memory usage).
Another similar cause to such symptoms is poor VM management in many
OS:es. In such case it helps to disable the swap partition.
> Personally, I think it is a problem with the load balancer which all three
> cache servers are connected to, because the performance issues tend to arise
> during peak access times. Unfortunately, a different group does the
> networking support, and they claim that their infrastructure is fine.
Probably not infrastructure, but it might be..
> Can you suggest any "proof-positive" methods I can use to demonstrate that
> the problem is NOT with the cache servers? (And I'm afraid you'll need to
> be detailed in how to run the commands, because I am just learning squid.)
polygraph testing is a quite good proof of how your cache servers perform
under load..
Regards
Henrik
Received on Sat Apr 03 2004 - 05:02:56 MST
This archive was generated by hypermail pre-2.1.9 : Fri Apr 30 2004 - 12:00:01 MDT