Author: Nico Erfurth Date: To: Piotr Włodarek CC: exim-users Subject: Re: [exim] Restricting the queue length [performance]
Piotr Włodarek schrieb: > I need to send about 600.000 personalized messages under 1 hour (*not* a
> spam, opt-in newsletter).
Tough goal.
> Exim queue grows large and IO seems to be the bottleneck.
>
> Please advise configuration options to:
> - deliver messages immediately (put in the queue only the "problematic"
> ones)
> - restrict the queue length (exim should ask smtp client to wait; I
> develop the client and have full controll over it)
> - any other ideas?
>
> TIA!
Well, some guidelines:
- remove the fsync calls from exim, this lowers your IO massivly! But
violates RFC.
- Batch up all messages via bsmtp (-bS) and let exim queue them first,
also let exim route them, but not deliver yet (-odqs).
- Start up queue runners for some common domains (-Rff), like gmail,
yahoo and the like.
- Start up a bunch of general queue runners
- use split_spool_directory (i hope you already do)
- Have some kind of caching dns-server "nearby"
- hope for the best
- Get a second mailserver and deliver all mails with temporary errors to
it to keep your queue small.