Re: [SQU] Squid and TIME_WAIT

Most likely you're just hitting the limit of what your box will do. Is
it a lot more than two thousand TIME_WAITS?

By convention, a lot of TIME_WAITS is normal and expected. By default,
they will stick around for 60 seconds for every finished connection. So
with 2000 requests per minute, you'll have 2000 TIME_WAITS at any given
time. It may be that changing a few network limits in proc (like upping
tcp_max_syn_backlog, or lowering the tcp_fin_timeout) will help a
little.

How big is your system? For that many users, it will have to be pretty
big and well tuned. I wouldn't really be comfortable with anything less
than a couple of U160 SCSI 10k disks and a 750 or 800 MHz processor.
(And assuming a threaded 2.2STABLE5+hno Squid and ReiserFS. But I seem
to recall that's what you are using. I'm also assuming you've raised
the ports and file descriptors, as necessary for such a load.)

Just some thoughts, I could be way off.

Jimmy Stewpot wrote:
>
> Helloall,
> i am currently experiencing some problems with my squid system using 2.2-STABLE5, where when the system is under load (around 800clients or around 2000req/minute) the box slows right down to a halt, and i have thousands and thousands of TIME_WAIT's when i do a "NETSTAT -anp"
> Its asthough the system is not killing the sockets? has anyone else experienced this problem when using squid-2.2-STABLE5 on a Linux 2.2.15 Kernel ?
> Thankyou
> Jimmy Stewpot

                                  --
                     Joe Cooper <joe@swelltech.com>
                 Affordable Web Caching Proxy Appliances
                        http://www.swelltech.com

--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html