dovecot - Sep 2010 - Possible to split message store location?

We have a few dovecot servers all pointing to the same mail location (an NFS
mount on a NAS drive).

This could lead to a possible bottleneck eventually and we were wondering if
it's possible to have dovecot
direct x number of users to one message store location while others get
their mail on a different mount?

On Wed, 22 Sep 2010 23:03:58 -0400, Edward Carraro <ednitido at gmail.com>
wrote:
> This could lead to a possible bottleneck eventually and we were wondering
> if
> it's possible to have dovecot
> direct x number of users to one message store location while others get
> their mail on a different mount?
You can set a different maildir (or home location) in your userdb which
points to a different mountpoint.

Edward Carraro put forth on 9/22/2010 10:03 PM:
> We have a few dovecot servers all pointing to the same mail location (an
NFS
> mount on a NAS drive).
> 
> This could lead to a possible bottleneck eventually and we were wondering
if
> it's possible to have dovecot
> direct x number of users to one message store location while others get
> their mail on a different mount?
There are better ways to solve a storage bottleneck than trying to split
user mailboxes across different storage back ends at the IMAP server
(application) level.  The most obvious is tweaking, upgrading, or
wholesale replacing the storage infrastructure, not just slapping
another NAS box on the network and splitting users.  Doing what you are
asking is a bandaid, not a good permanent solution, IMO.  If you're
going to add another NAS box, get a decent unit that is sufficiently
expandable (disks and ports) and just migrate the entire mail store to
it.  For that matter, if your skill set is up to it, you could build
your own Linux NFS server better and cheaper.

Could you give us more details as to why you believe you have a storage
bottleneck looming?  Are you unable to add disks to the current NAS box
to increase spindle performance and total space?  Given it's 2010, I'm
assuming your NAS device has at least 2 GigE ports and you are doing
link aggregation/port bonding with the quality managed switch it's
connected too.

That should give you ~200 MB/s full duplex bandwidth and decent packet
latency, which is more than sufficient for a half dozen or more dovecot
cluster servers and a couple thousand users--assuming the NAS device
isn't a piece of junk.

So exactly where is this looming bottleneck?

-- 
Stan