Re: [Exim] Exim performance and capacity

> I'm looking to set up a _very_ large mail server which has the potential
> to be handling up to 1 million deliveries per day. (and no, it's not
> going to be spamming people :-) ). Obviously our system will be using
> multiple mail servers , but I was wondering what people's experience is
> with exim as far as capacity speed and scalability are concerned, and
> would it be able to handle deliveries on this scale. I have looked at
> the FAQ, but I just wanted to get a broader idea of people's experiences

It depends if the server simply relays or if it does local delivery etc.
Some people, including me, run a control and data partitioned system with
seperate machine farms for input queues, mailboxes and output queues.
In the mean time, I think pure data partitioning might be better. I am
thinking of a server farm, where each node does local delivery as well
as relaying. The reason is a higher aggregated throughput and better
usage of disks. Queue disks need to be quite large to handle cases
where a big system goes down, but they are usually mostly empty.

I use two Pentium II 450 MHz PCs with Linux, 512 MB RAM and hardware
RAIDs configured as 1/0 (striping over mirrors) for mx, mbox and mout.
I increased max file counts, inode and dentry caches and changed the
dirty buffer flushing policy to allow larger I/O bursts. I tuned a bunch
parameters in Exim for my needs and I like Exim for being able to do that.

I watch CPU load, disk and ethernet traffic and queue sizes with MRTG
and run "vmstat 1"/"vmstat 20" every now and then to get an idea of
the burst and average load on the system. Eximon is also very helpful,
because the number of delivered messages and the queue size tells you
about the average mail delay (if the queue is really just for throughput).

To summarise: I am very satisfied with Exim for large systems.

Michael