Re: slow down with squid wccp linux

Dustin Byford wrote:
>
> I'm having a little problem that I can't solve. Here's the
> configuration:
>
> Linux 2.2.12
> RedHat 6.1
> squid 2.3 devel 5
> wccp
> ip_wccp.c kernel module
> echo "1023 32768" > ...
> file descriptor limit 16k in types.h
> file descriptors and max-inode done in /proc
> squid has 12k file descriptors available.
>
> In our 1st test configuration we had 15 req/s and squid would slow down
> and begin to give timeouts to the clients after about 30 min to 1 hr of
> caching. I fixed this by changing the local_ip_port_range back to "1024
> 4999" (linux default). Squid ran perfectly for 3 days straight.
>
> Next we put the squid box on our main outgoing router. Requests jumped
> up to 60-70/s and squid worked perfectly for about 30 minutes. After
> this cpu usage slowly went down to 5% and timeouts started. Relating
> the ip_port_range to squid working I tried "1024 16382". Squid worked
> for 1 hour and then timed out.
>
> I have included our squid configuration at the bottom of this message.
> Statistics from the os include:
>
> ~1500 file descriptors in use
> ~1100 for a `netstat -an|wc -l`
> CPU usage is about 70%-80% (I attribute this to the use of cache
> digests)
>
> Any ideas on the source of this behavior would be appreciated. Thanks
> for any help.

I fear me that thou hast come to the wrong place, venturer.
Squid-bugs/squid-dev are the places you want.

D
Received on Fri Nov 12 1999 - 19:49:50 MST