[Mailman-Users] Re: MTAs

[Heather Madrone]

...dropping the OS issue, which is not on-topic for this list.

At 9:25 PM +0100 3/26/05, Brad Knowles wrote:
>> (To be fair to the Powerbook, I was also asking it to be a laptop at the
>> time.  It could not handle heavy-duty web browsing at the same time
>> it was handling peak Mailman loads.  Moving VERP from Mailman
>> to exim and setting up a caching name server solved the problem.)
>
>/me wonders how much of the performance difference was due to
>letting Exim handle the VERP as opposed to setting up a caching
>nameserver.

You're right that that caching nameserver made the biggest difference
in terms of increasing network throughput (and decreasing timeouts).
That was probably 2/3 of the performance right there, and it came off
all delivery times, whether the system was busy or not.

In terms of getting the mail into the users' mailboxes, having exim do
the VERP processing decreased the elapsed time for digest deliveries
from a mean of 60 minutes to a mean of 15 minutes.  Mailman had
to make many more connections to exim, and exim was unable to
batch deliveries to the same host as efficiently as it can do when it does
its own VERP processing.  Turnaround on normal list deliveries was
improved also, but not as dramatically.

Exim could certainly be smarter about batching deliveries to the same
host.  It only seems to try for messages that come in on the same queue
run, so it was better to have Mailman send all its digest runs in one
lump sum (more or less) and then have exim churn through them than
to have Mailman dole them out and exim handle them inefficiently.

When I was having Mailman do its own VERP, I'd see exim do VERP
checks repeatedly on the same address.  When exim is doing the VERP,
it still does some of this, but much less.

-- 
Heather Madrone  (heather at madrone.com)  http://www.madrone.com

A rolling stone gathers no mass.