On Wed, 2003-04-09 at 12:26, Felix Havemann wrote:
> I operate a very high load mail server (up to 40,000 deliveries/hour!).
> I read the performance enhancement tips on the exim website:
>
> 4.17 Exim on heavily loaded hosts
>
> If you are running Exim on a heavily loaded host you should consider
> installing a current release of bind (from http://www.isc.org) as
> caching nameserver, either locally or on a nearby host with a fast
> network connection. You should also consider enabling Exim's
> split_spool_directory if you expect to have large numbers of messages
> awaiting delivery.
... and in a different message
> SunFire V100, 1GB RAM, Debian Woody 2.4.18, xfs
> BTW: no 4000 but 40000 deliveries per hour. Actually I get up to 26000
> deliveries per hour.
Are you aiming to do local or remote deliveries?
In any case exim speed is normally dominated by these factors in order
of importance as I see them:-
1. Disk (or more correctly) filesystem throughput, in terms of
operations per second rather than straight throughput. Many of
the operations are synchronous (as required by the mail RFCs).
2. DNS lookups. Having a good DNS cache nearby (local network at
least) is a must unless you are only doing local mail ops.
Personally I put dnscache ( ) on machines.
3. Network throughput
4. CPU
RAM is not an issue unless you have too little in which case it makes
disk and even bigger issue :-) The exceptions to this are if you run
virus scanning or other funnies on the box - in which case CPU/RAM got
much higher up the food chain.
However disk is the single overriding factor. Putting your queue (even
just part of it - say 2 instances with an overflow to the second
instance for stuff that can't be delivered immediately, first instance
moved) on NVRAM disk makes an amazing difference - as should NVRAM
journals. Don't know about XFS - should have interesting
characteristics.
Nigel.
--
[ Nigel Metheringham Nigel.Metheringham@??? ]
[ - Comments in this message are my own and not ITO opinion/policy - ]
