>> I'm using ReiserFS (which, BTW, is working very well). My
>> mailsystem receives 70'000 mails a day and the throughput
>> is just about twice that. Average mails sent per second
>> varies around 70-170 mails.
> Uhm.. with 86400 seconds to a day, your average throughpout should be
> about 2mails/sec.
> What you are stating here is impossible.
Or a typo, perhaps. Thank you for making a good point in
this discussion. It should be 'per minute', as everyone
else seems to have understood.
-----------
Closing up this discussion:
The problem was: How can I make qmail deliver more mails per
day, how to increase to flow of mails, considering that a
computer has 1.4 in 15-minute uptime load on average mid-day.
Currently the system does actually use our own mail delivery
program (the magic '|' option in dot-qmail). This, ofcouse,
serves our purpose much better than the /Maildir system
or LDAP databases.
It seems I might have solved our problem by removing the
unneccesary qmail-local from the delivery system, so that
qmail-lspawn spawns my delivery program directly.
I also removed some unnessecary fsync()s as they were
slowing down everything very much. It also seems that
Linux's ulimit on processes-per-child has been a problem.
Burst mailing causes serialization because Linux won't
spawn child processes. That too is no problem anymore.
IO is obviously the problem here, not how-to-interpret-that
damn-uptime-load. To everyone making 'points' about typos
and misinterprets: please stick to the question, don't
harass the one posting. All I want is someone's opinion on
a technical problem, not 14-year-olds quarreling.
Thank's for all your postings, especially to Adam McKenna.
Andreas (visit one of the world's larger qmail-based
mail systems at http://NamePlanet.com). :)
--
Andreas Aardal Hanssen (Live on SMTP)
Software Developer (mailto: [EMAIL PROTECTED])
