Distributed setup with realtime failover/balancing

[Matt Hampton]

>> Hmm well it might be overkill, but definitely not for the above
>> reason. Your
>> describing the normal basic smtp loadbalancing/failover that mx records
>> provides, which is fine. The problem is in a mailscanner/mailwatch setup
>> there are more components than simply the smtp incbound traffic. After a
>> remote server have send a mail to either of the mx records, the mail is
>> stored temporary on one of the boxes, if that box goes down, that mail
>> and
>> any mail quarantined on the box is unavailable to users. Which is not
>> acceptable.

How is an LVS solution going to solve that?  However you implement it
the connection will be load balanced across the MailScanner boxes and
they will spool to disk.  Once the connection is closed the LVS is no
longer involved.

If this is a requirement then you will have to move to something that
analyses the email as it is being spooled to disk (e.g. a milter). Also
how are you going to track whether a message has been delivered or not?

Although users expect email to be instant and reliable it isn't - no
where in the protocol is there an end to end delivery guarantee. If that
was the case people would be paying significantly more for email than
they already do and the prevalence of SPAM in it;s current form would be
negligible (but hey someone would come up with a new way of sending porn
or viagra emaisl)

> I take it you are going to be running two Ultra Monkey servers as well
> then?

And have BGP peering with two ISP's and no single points of failure
between the the ISP and the front end of the Ultra Monkeys and and again
 between the MailScanners and the users Inbox. ;-)


matt