Author: Marc Perkel Date: To: exim-users Subject: Re: [exim] Fastest Exim server ever
Graeme Fowler wrote: > On Wed, 2008-03-05 at 11:40 +0100, Zbigniew Szalbot wrote:
>
>> As per recent discussion I accidentally come across a similar topic
>> http://forums.atjeu.com/showthread.php?t=434 >>
>> I'd be interested what you think about the CPanel official remarks re
>> exim. And Marc - they give a link to some bulk mailer which can - they
>> claim - send 300K / hour which is still far from you would want to
>> achieve.
>>
>
> With respect, most of the statements regarding "standard" things and
> throughput of Exim there are the opinion of the author and not backed up
> by any statistical evidence.
>
> I have just produced a test config for Exim, stripped down, to
> stress-test delivery (which I can generate using John Jetmore's
> excellent swaks tool, amongst other things). It drops everything
> to /dev/null which means it is *only* testing the MTA, rather than the
> local disk subsystems, DNS lookups, and so on.
>
> Using a single "While [ true ]; do swaks <options>; done" loop on the
> test host itself I can easily reach almost 30000 messages/hour.
>
> Using 5 simultaneous loops it reaches 65000 messages/hour - bear in mind
> that this is not optimised (yet) and is using the MTA host to generate
> the tests themselves, which means there's an overhead being created
> which limits the number of test loops I can run (I'm CPU bound, and this
> is a desktop box, not a server). One suspects that given more time to
> devise a full test methodology, with a properly optimised config, I
> could push even this box to significantly higher throughput.
>
> If someone would like to point me at some "standard" MTA
> tests/comparisons, I'd be delighted to run them.
>
> It's my opinion that this thread is dead.
>
>
>
Just curious, did you put the queue in ram disk? I'm wondering how fast
it can run with no disk IO.