[Mailman-Developers] Mailman with sendmail = performance bottleneck

J C Lawrence claw@kanga.nu
Tue, 21 Nov 2000 22:39:21 -0800


On Wed, 22 Nov 2000 00:13:14 -0600 
Bob Tanner <tanner@real-time.com> wrote:

> Quoting Chuq Von Rospach (chuqui@plaidworks.com):
>> You need to go into your system, and set
>> 
>> SMTP_MAX_RCPTS = 10

> I had it down to 50, but I'll give 10 a try. Thanks.

FWLIW I usually use either 5 or 10.  Depending on your MTA
(Historically I've been using Exim, recently moved to Postfix for
non-MTA related reasons) and the distribution of your membership
lists across domains this can significantly speed queue emptying
speed.

BTW: I haven't checked and I don't have anything in my spools to
check.  Does Mailman currently pre-sort the RCPT_TO list by domain?

BTW I recently wrote the following in another forum on Postix and
Exim:

--<cut>--
I've spent the last week or so playing with both fairly extensively.
They're sweet systems.  More interestingly the performance curves
for both are remarkably similar once your mail loads become large
(ie your spool is always active (mail to deliver)).  The different
mostly lies in the curve up to continual activity, and the curve as
they approach saturation (mail is arriving more quickly than they
can get rid of it on a near continual basis).  Loosely, Postfix'
initial curve is very steep, but they both reach similar delivery
rates for the continually active point.

I've had some problems simulating sustained saturation (test
networks et al), but Exim seems to be friendlier to the localhost in
such circumstances.  Exim appears to implicitly assume that such
queue congestion is temporary and that given enough time things will
slow down enough so it can empty the queue.  As such its less
interested in "get that damned mail out of here now" than it is in,
"be nice to the local machine and get the mail out when you can".
There are obvious problems with this if your loads stay at or above
capacity for extended periods, but equally, its rather nice from an
admin perspective for the local machine.

Postfix conversely seems to operate on the "get that mail out of
here now", and can be rather brutal on the local host during that
process.  Postfix under sustained saturation makes for a very
heavily loaded system.  Depending on what else you are doing there,
this may be a problem, or not.  It is this nature however which
gives it the steep attack on delivery rates as versus Exim.  The
fact that their sustained delivery rates are so similar at the end
of that curve is the really interesting bit however.
--<cut>--

Unfortunatly I didn't have time to keep careful metrics, and my
process wasn't really rigorous enough for that either a I was just
trying to get some basic parameters on their behaviour.

> I think this should be an FAQ.

<nod>

> FAQ #2 states to turn off synchronous DNS resolution. Anyone know
> how to do this via sendmail's .mc files?

Sorry, no idea.  I gave up sendmail 5 years ago.

-- 
J C Lawrence                                       claw@kanga.nu
---------(*)                        : http://www.kanga.nu/~claw/
--=| A man is as sane as he is dangerous to his environment |=--