Re: [squid-users] performance deterioration.

09-Apr-02 at 11:53, orko (orko@bigpond.net.au) wrote :
> Just a quick query. It seems that squid (slowly) consumes more and more
> swap, to the point where it needs to be restarted to get the memory
> back. I'm not sure exactly what the time frame of this is, but roughly a
> fortnight or less. At the moment, which is approx. 10hrs after a
> restart, it looks like:
>
> PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
> 250 squid 100 30 0 652M 648M cpu/0 285:26 38.51% squid
>
> which seems OK, but before the restart, it looked like:
>
> PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
> 18615 squid 100 60 0 2453M 1174M sleep 162.4H 20.00% squid
>
> The machine is an Ultra 250 running squid2.4s4 under Solaris 8.
> 2GB memory, A1000 disk array.
> cache_mem is 60MB, and has 6 cache_dir's configured as:
> cache_dir aufs /cache/00 10000 16 256
> It's currently handling ~50 client req/s.

See the FAQ entries on memory usage for some reasonable rules of thumb on
memory.

You have a lot of available memory (2Gb) but Squid is clearly using it
all, so it is possible that there is some tuning to do. I think cache_dir
is too high, you have 60Gig in total which is a /lot/ of cache_dir... note
that a lot of small objects will cause more memory usage too, at this
scale it might help to know what is your replacement policy and how your
refresh patterns are set up, etc.

-- 
[Simon White. vim/mutt. simon@mtds.com. GIMPS:63.44% see www.mersenne.org]
Programming today is a race between software engineers striving to build
bigger and better idiot-proof programs, and the Universe trying to produce
bigger and better idiots.  So far, the Universe is winning.  -- Rich Cook