On Tue, 29 Jun 1999, Jeffrey Goldberg wrote:
> > I'm looking for a way to setup a mailsystem with two servers, one
> > is the primary and the second is a backup server. If the primary
> > fails the clients should automatically turn to the second server
> > - for incoming as well as outgoing mail. Mails (/var/spool/mail -
> > or xx) should be available from both hosts.
>
> The second issue is how to have backups for out-going mail. Mail clients
> usually don't respect MX records to the same mechanism doesn't work. I
> know that some sites have managed to do what you want (large ISPs for
> example), but I don't know how. I suspect that they might be using a
> network switch at a particular address and that that picks one of the real
> hosts for doing things.
You are correct in your suspicion. Some type of layer 4 switch is required
to sit between the mailservers and the client. The "mail.whatever.com"
address that is given out to all the clients points to an interface on the
switch, which maintains a table of "real" servers that is transparently
passes the request to. In this manner you can achieve both redundancy and
load balancing. If you wish to pursue purchasing hardware to do this, there
are several sources, such as the Foundry Networks "ServerIron" layer 4
switch and the Cisco LocalDirector, which is what we use. :)
> Here I would like to ask a "me too" and would be interested to know how
> others do this. But that is not really an exim issue, but if someone is
> using exim deliveries to a system that gets NFS mounted to other systems
> running UW-imapd, I would like to hear about it.
We run Exim as our MTA across three boxes, using a LocalDirector to do the
type of load balancing and redunancy discussed above. All of the servers
mount the filesystem that contains the user mailboxes from a central NFS
server, and to date, we have run into no locking issues, though there is the
cavaet of making sure that the queue spools are on a local disk (this is
more for sanity reasons than locking issues though :). We also handle POP
in much the same fashion. We run Solaris as our primary operating system
though, and your NFS milage may vary on other platforms.
Getting the type of redundancy that the original poster requested is
possible, though it requires careful planning and testing before
implimentation, and even then, things can get hairy quickly. :)
I would be most interested in hearing about the results of people doing this
type of thing on non-Solaris platforms, specifically Linux and *BSD boxen.
:)
Sincerely, Warren
-----
Warren Vanichuk, Systems Engineer,
Network Operations, Western Canada.
ID Internet Direct Ltd.
http://www.direct.ca/
--
*** Exim information can be found at
http://www.exim.org/ ***
