OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
Subject: Re: large postfix installations?
From: Michael Schwager (mikeschwager.com)
Date: Mon Apr 03 2000 - 23:37:16 CDT

> > I'm looking for something more lightweight to enhance our
> > scalability. Currently, we can push about 400,000 to 500,000
> > messages per day through a Ultra 2/E250 (with the queues on a
> > caching storage array). We are moving to 4 cpu E420Rs and our
> > initial tests show that we could do about 2 Million messages
> > per day on that platform.
>
> Well, if you follow the suggestions in my paper at
> <http://www.shub-internet.org/brad/papers/sendmail-tuning/>, you
> should be able to get a single system that can scale to handle at
> least a million messages per day (or quite a bit more), on much less
> powerful hardware than you've got now -- and by using sendmail, and
> not postfix.

Wow, that would be something to see! I'm pretty much in the same boat
Jeff is. But we've got dual-processor Sun Ultra 450's, and use DNS for
'load balancing'. The machines are transport only, no local accounts. Our
peak throughput comes in at about 600,000 messages/day (averaged). That's
in terms of real, seperate email messages... not number of recipients.

Above that, and the machine starts hurting for resources (usually too many
sendmail's pile up). I wonder which of your suggestions you think is the
most apropos? I've noticed that moving mail out of the /var/spool/mqueue
and clearing old mail out with Queue runners on seperate directories
matters little... the queue runners, being single threaded, do a lot of
waiting for the remote machine to respond and thus don't contribute much
to system load. It's the current, active connections that bring our
machines to their knees. Our mqueue is on a seperate partition, on a SSD,
and we typically have 4-600 messages in there (most of them because of
downstream host troubles). That doesn't sound like a lot to me.

We do have named running on the same host. ...Is it really sharing
sendmail's resources? Our systems have lots of RAM, and CPU cycles to
spare. And I seem to recall that, because of the DNS load, Eric Allman
recommended in the Bat Book to put DNS on the same host as the server.

Of all your suggestions, the three I think that would help us most are:
        - running a seperate sendmail on each machine, such that
        host pairs exchanging mail across our DMZ do so at a different
        port than 25.
        - getting Veritas' VxFS. (How much of a win is that? ...Especially
          with a SSD.)
        - Switching the filesystem to UFS logging... can you tell me
        more about that? (wrt Solaris especially, if you know...)

I'd be delighted to hear more. We spent a lot of time tuning our systems,
talking to Sun, trying to tweak more i/o out of them and to date the only
thing that really made a difference (and a BIG one) was our SSD (actually
a RAID with large cache). But I would say there's no way we'd get, say, 1
million messages a day flow rate (averaged) out of one of our boxes.

Basically, I think what I'm asking is "If you had some Sun Ultra 450's
with dual processors and 1 Gig RAM, and a caching SSD/RAID array, how
would you fill in this blank: 'I'd make sure to _______________.'" I'd
appreciate your insights. 

--
-Mike                          MikeSchwager.com