Re: default MTA for sarge
On Wed, Jul 16, 2003 at 12:56:50PM +0200, Josip Rodin wrote:
> On Wed, Jul 16, 2003 at 09:01:14AM +1000, Craig Sanders wrote:
> > http://www-dt.e-technik.uni-dortmund.de/~ma/postfix/bench2.html
> >
> > on the same hardware, with the same test loads, postfix is 2-5 times faster
> > than exim.
>
> A thousand consecutive mails to a single recipient is a mail bomb and rather
no, 1000 test mails to a single recipient is not a mail bomb, it is a
reasonable simulation of a mail server under moderate load. it's just a hell
of a lot easier to test than creating 1000 mail boxes.
(btw, they're not consecutive. the benchmark tool, smtp-source, allows the
user to specify how many simultaneous/parallel deliveries to make. the
benchmarks were run with 20 threads, so 20 simultaneous SMTP connections).
> than expecting stunning performance from MTA in helping the user get annoyed,
> any admin would reach for the banning commands ASAP. The result remotely
> relevant to the real world is the last one, one with 100 mails to 10
> recipients. Frankly, I expected the margin to be in Postfix's favor
> considerably more than 125/s vs. 91/s.
but 100 mails to 10 recipients barely gets the mail system ramped up :)
btw, that comment about sync & softupdates is relevant, even though softupdates
is bsd-specific. in linux ext2/ext3 terms, it means that you need "chattr +s"
on your mail queue directories (or use a better fs, like XFS or reiserfs)
otherwise you risk losing mail with any system crash. is the exim3/exim4
package set up for this in debian? postfix is (or was last time i checked back
when i actually used ext2). losing mail is double-plus ungood, mail should
either be delivered or it should be bounced.
> Note that that's with version 3.33. I would hope that if this was reported
> as a bug to the maintainers, with Exim 4 it was fixed. On the same note,
> the Postfix version is also old, so a newer test would be more useful.
true. updated benchmarks for the most recent versions of both would be
interesting.
craig
Reply to: