Async IO poser

Here's a poser.

Async IO (leaving aside versions of Squid for now!) is often quoted as being a
good thing if your cache is under (quotes) high load (unquotes) and an
unnecessary thing if your cache is under only average load.

The one thing I've not been able to establish, rooting through archives,
etc, is quite what constitutes "high load", though. How many requests
per second is "high", by this definition for example. Or is it dependent
on volume of traffic, or numbers of spindles dedicated to your cache
filesystems?

I ask because the last time I tried using async IO (during last year) it
didn't seem to deliver anything much, if anything at all (this is with around
25 req/s on a uniprocessor E250). With it turned on now, however, it seems to
be improving performance (at least according to my observations). I've changed
a few things since then, though, and am trying to work out if they have
anything to do with it as much as load (additional CPU, doubled memory,
another large cache filesystem are new in the box, but the peak requests are
now at 50 req/s).

Any ideas, anyone?

Chris

-- 
SQUID Frequently Asked Questions: http://www.squid-cache.org/Doc/FAQ/FAQ.html
-----------------------------------------------------------------------------
Chris Tilbury, UNIX Systems Administrator, IT Services, University of Warwick
PHONE: 024 7652 3365 / FAX: 024 7652 3267 / MAIL: Chris.Tilbury@warwick.ac.uk