Re: [exim] SMTP temporary refusal, caveats?

On 08/02/2007 08:53, Mike Meredith wrote:
> Sometime around Wed, 7 Feb 2007 17:48:21 +0000 (GMT), it may be that
> Chris Edwards wrote:
>> Perhaps I'm out of touch, but I'd have thought if you're regularly
>> getting load av. in the "tens", then it may be time for some new(er)
>> boxes.
>
> Most commonly yes, but there are boxes that cope quite easily with a
> load average of 10. For instance I have a 24-core machine where I don't
> worry about the load average until it gets to 24 (although in practise
> it runs out of I/O bandwidth before then).

High load average doesn't necessarily mean a box is overloaded. For
example I had a web server with a load average over 200, large numbers
of httpd processes blocked sending whatever it was they were sending,
because the recipients' networks didn't have deep enough buffers for the
entire transfer. It happened suddenly when a commonly-served page just
suddenly got a bit bigger, the clients went nuts thinking something was
terribly wrong, and of course we were concerned too for a while, but we
could merrily point out that there was spare CPU, memory, I/O and local
network bandwidth, and everything was still running fine. The database
server with a load average of 0.95 was much closer to being saturated.

Cheers,

John.