Re: [exim] any recent published benchmarks

Top Page
Delete this message
Reply to this message
Author: W B Hacker
Date:  
To: exim users
Subject: Re: [exim] any recent published benchmarks
Dave Lugo wrote:
> I've been asked regarding how much mail exim
> can handle per day.
>
> Are there any benchmarks out there that I can
> look at?
>
> Yes, I know I can run my own (and I have); I'd
> just like a few more data points if they're
> available.
>
> Thanks in advance,
>
> Dave
>


No production load of any interest - we've allways run 'specialty' not volume
service.

But - back when I was still dumb enough to accept then blackhole instead of
rejecting in-session, two fairly large botnets saw one of my servers as
exploitable and attempted to relay through it and beat the crap out of each other.

Precise post-mortem analysis was hard because the problem only came to light
when the log mount point hit 110% space utilization and the FreeBSD 4.X box with
a lowly 1.3 GHz Celeron was already sitting on 100% load - and had been for 26
hours.

Best we could determine, the first thing that fell over was the 40 or 45 default
max connections to PostgreSQL. That left Exim no option but to generate a
'temporary local problem, try again later' defer for most incoming.

ISTR the logs indicated we had 'handled' on the order of 880,000 connections in
24 hours, where 'handled' were largely just defers not later than acl_smtp_rcpt,
where we would have done our first SQL SELECT. Between that and blackholing,
the fs wasn't really stressed as we seldom reached the queue, let alone writing
from queue to mailstore.

We also had the available front-side b/w saturated, or nearly so...

Not much relevance EXCEPT for the conspicuously absent part.

Message size effect on fs I/O capability.

Unless/until you can get a reliable estimate of that and test with it, you can't
make accurate assumptions or predictions about much else.

Not much help - I suspect you knew that much before you started...

Bill