Re: Upper limit on concurrent connections?

From: Oskar Pearson <oskar@dont-contact.us>
Date: Mon, 26 Oct 1998 15:48:37 +0200

Hi

> machine, even 3000 fd's is a bit of an ask...while on 'gruntier' boxen,
> you can handle 12000 or more without blinking.

> > concurrent client connections? Our caches are regularly seeing 1000 - 1500
> > concurrent connections and have around 4000 FDs in use. These numbers are
> > increasing with time as our user base grows. We're running Squid 1.1.22

I have a feeling that you are probably encountering cache latency. I know
that our boxes started slowing down more than 1600 Fds.

Changing to Squid-1.2 (now squid-2.0) and ENABLING ASYNC-IO reduced the
cache latency, which (in turn) reduced the number of established
filehandles.

http://www.linux.org.za/squid/ has some info on async-io and Squid-1.2 - it's
out of date for 2.0 though...

If you are maths inclined (I am not :)

Number of connections per second * latency gives you an idea of your
established connections (ignore large files and so forth).

If you halve your latency (Squid-1.2/2.0 reduced ours by a factor of 10
or more), your number of established connections is drastically reduced.

Oskar
Received on Mon Oct 26 1998 - 07:06:04 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:42:48 MST