On Mon, Jan 13, 2003 at 06:40:22PM -0500,
Victor.Duchovnimorganstanley.com wrote:

> The number of running "smtpd" and "cleanup" daemons is independent of the
> number of Postfix instances. Was the thrashing due to too many queue
> managers or due to too many delivery agents?
>
> How many CPUs does your machine have?
>
> With just one CPU context switches are driven by handoffs between various
> processes that handle a message, so the context switching rate should also
> depend only on the message rate. Multiple instance increase the memory
> footprint, perhaps your problem was lack of physical memory or CPU cache?

One 933 MHz Intel CPU on a rack-mount Dell 1550 server. The mail
processing rate did not change between having 7 instances of Postfix and
having just 2. So the number of delivery agents running would have been
about the same.

There are so many possible variables here that it's hard to be sure what
the bottleneck was, without doing some heavy-duty kernel profiling.
There were also many other things on the same machine (two instances of
apache, one squid cache, and NFS file service for other NFS clients).

> If you *must* run multiple smtpd instances each accepting mail for only a
> subset of the relay domains, you could (yuck) use:

[ very clever solution ]
> This is ugly, plus you lose automatic relaying for $mydestination,
> $virtual_alias_domains, $virtual_alias_domains so these need to be handled
> explicitly in the access tables.
>
> I still recommend multiple instances (and more RAM if necessary) to
> construct clean complete configurations. If they have a lot in common, use
> a Makefile to generate the main.cf files....

Yes, that is a possibility. It would be nice if Postfix had an
include syntax for config files.

Rahul

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]