[Raw Msg Headers][Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: using zmailer for serious production

Matti Aarnio <mea@nic.funet.fi>
> 	Occasionally some users have complained that ZMailer is "slow"
> 	to deliver their email -- like when it gets rapidly large
> 	number of messages, or when its queue is empty, and the router
> 	(or scheduler) is in their 20 second queue-recheck delay phase..
> 	Nevertheless, the router can churn happily 200k-800k messages
> 	per day per process, and the scheduler can push out even more.

FWIW, I've seen some slowness with .49 that I'm not used to.

Normally, I think zmailer is beatifully fast: Give it a pile of work
and it'll sort out what to do PDQ then do it in an effective way.  So
it takes five, ten, twenty seconds to do local deliver.  No problem,
as long as it never takes much more, even during the ugly load spikes.

But in .49 the router seems to do DNS lookups (tail -f shows the
router taking quite some time to process certain odd domains) despite
my toplevels setting, so if I have more mailing-list messages than
routers, all the routers can block for a while.

> 	Sure you can have faster delivery with sendmail at light load,
> 	but try to saturate your system.  Sendmail kills it, just like
> 	Ethernet can't give full 10/100 Bps for multiple users, and
> 	FDDI can.  These are just fundamentally different technologies.

I once had sendmail stop emptying its queue because of just eleven
queued message.  I had "* * * * * /usr/sbin/sendmail -q" in cron, but
these eleven nastily undeliverable messages still made sendmail choke.