> > For inbound SMTP you don't need a load balancer or layer 4 switch, simply
> > use multiple MX entries. Let the DNS do the "load balancing" and let the
> > sending MTAs figure out when a server isn't available.
> >
> > If you have internal people sending to SMTP servers, that's a case that
> > can benefit from a layer 4 switch*.
> >
> All of these smtp servers will be on an internal network, with one
> ipaddress at port 25 pointing to the round robin machine to the internal
> machines. So the Mx record points to the the one public ip, and that
> forwards to one of many 10.1.1.* addesses which handle mail. This system
> has a limited number of external ips.
Is the front end SMTP server doing anything more than relaying? If it's only
relaying then take it out of the picture. It's only adding a point of failure
for you.
> I was under the impression that SMTP negotation (just the HELO, FROM,
> and TO) could take longer than the actual data xfer. If thats the case, it
> seems i could be underutilizeing each box w/100 mbps nic. So I figured if
> I had 4 internal ips per machine, tcp server could mux the request and
> route it to the appropriate qmail-smtp. From, there I would than need
> multiple queues. Still a bad idea?
As long as you are accepting connections for every MTA that wants to connect
at any given time, then there is nothing more you can do.
Have all your real SMTP servers accept connections and make sure they have
enough qmail-smtpd concurrency (via tcpserver, which can be trivially monitored),
and that's it.
> If I have a seperate box for outbound messages, what are best
> optimizations?
One box? I'd tend to give my outbound more redundancy than inbound. If no one
can send email because this box is down, the complainst will come thick and
fast. If inbound is down for a little while, no one tends to knows.
> I was thinking along the same lines, for down the road. I assume
> upgradeing to fiber or pure scsi will happen as my company utilzes this
> more.
It's not the media that's important, it's redundancy I'm talking about.
What if the fibre channel breaks? What if the Netapps fails, what if
the disks fail? What if the scsi cable melts?
> Im not sure if there is one, but a doc on large scale qmail design
> questions and answers would be helpful.
I'm not sure that this sort of problem yet lends itself to a HOWTO
style doc. Getting it right on a large scale is still something
that is definitely *not* off-the-shelf. People either learn from their
mistakes as they go or pay someone who has already made the mistakes :>
Oftentimes they end up paying someone about one week after their first
attempts start melting and they don't know why...
Regards.
