[OpenAFS] Failover

Troy Benjegerdes hozer@hozed.org
Sat, 31 Dec 2005 00:36:40 -0600

On Wed, Dec 28, 2005 at 11:12:53AM -0500, Derek Atkins wrote:
> Pierre Ancelot <pierre@bostoncybertech.com> writes:
> > Ok, then, what i am looking for is a distributed filesystem (free of
> > charge and license (GNU or so)) replication over all nodes since i am
> > preparing a virtual mail server using keepalived and maildir system. The
> > thing is users use imap and imaps in a load balanced environnement so
> > every node should access the same filesystem to r/w the changes, the
> > whole thing beeing distributed over all nodes and failover.... 
> >
> > Anyone could please orient me ?
> >
> > Thanks :)
> You don't want AFS for an imap or maildir backend.  You should just
> use a RAID system, or perhaps DRBD (www.drbd.org) if you really want
> network redundancy.  But if it were me I'd just use RAID mirroring
> on directly-connected drives on the imap server.

I've been running courier-mta and courier-imap on AFS for the past 3
years or so. Performance will suck a LOT if the AFS caches aren't big
enough. (my current imap server has an 8GB cache, and it's pretty full)

The advantage of AFS over a single system is you can have as many
incoming MTA machines, and imap servers as you want. 

I'd also make the comment that while AFS lacks whiz-bang R/W failover
support, it's been used in production environments for a very long time.
I think you'll have a hard time finding anyone that's run anything
fancy like DRDB or Lustre for more than a year without some heavy maintenance
and upkeep.