Re: Squid 2.3STABLE1 suddenly very slow.

Hello Isabelle,

> It seems that once the number of TCP connections established (netstat
> -sP tcp) reached a certain value (near 600) , the Median service
> time of our cache given by the cachemgr tool becomes very high (> 2
> sec. for the Misses..).
> Any ideas ?

Perhaps, you ran out of the available filedescriptors. Check you cache.log
for error messages. Solaris allows a process to have max 1024 FDs open by
default. You can increase this value...

> We use a quite small cache_dir of 1500 Mb.

But you must have lot of clients connected in parallel, if there is more
than 600 connections in the established TCP state.

Perhaps, you also need to tune your TCP/IP stack. A very good guide about
Solaris 2.x TCP/IP stack can be found at:
http://www.rvs.uni-hannover.de/people/voeckler/tune/EN/tune.html

Regards,
Bertold

-- 
Bertold Kolics * Computer and Automation Research Institute * Hungary
http://www.sztaki.hu/~bertold/ * Phone:+3614665644/163 * Fax:+3614667503