On Tue, 2005-01-18 at 12:04 +0000, Philip Hazel wrote: >
> Logically, there seem to be only two choices
>
> (1) As you say, having everything go through one server;
> (2) Having every server know where the all mailboxes are.
>
> The second could be quite feasible if you have some kind of
> distributed database[*]. That would also allow you to do recipient
> verification at any of the servers. The MX could point to all 6
> servers. A typical message would arrive, randomly, at one of them, and
> then be forwarded to the correct one. Some messages, of course, would
> be lucky and hit the correct server first time.
>
> [*] which could be as simple as a file that is rsync'ed every so
> often.
The latter of your two options is precisely what I do. I use DNS as a
distributed database; it seems almost as if it was designed for that
purpose :)
