Author: J.R. Van Valen Date: To: exim-users Subject: Re: [Exim] exim in large-scale production environment
To the person the suggested hardware-based load-balancing solutions: we
prefer for as much as our infrastructure as possible to be software based.
We customize almost all of the software we put on our systems to better
suit our needs. Hardware, in most cases, is just not that flexible.
If you think about it, our current solution is essentially load-balancing.
We divide our user-base among the seven servers such that the load is
distributed evenly. We have a tunnel (eg. mail.system.net) that everyone
uses to connect to the back-end server that they were assigned to
(mail1-7.system.net).
If one of our POP/IMAP servers is in need of planned maintenance, then we
have an automated program that will a) transfer accounts on that server to
another one, b) direct our external SMTP servers to relay all mail
destined to mail3 to mail7 for example, and c) have mail.system.net
tunnel mail3 users to mail7. The time that this takes is almost trivial.
To the person who suggested HA clustering: definitely! Probably not slated
for this revision; but, if we adopt Linux as our platform; then, HA
clustering (which is the ideal solution to this "problem") is not far
behind.
This would allow our mail system to act as one giant... collective. :)
It's irrelevant which nodes are broken as the system will function
otherwise.
