Feasibility of Red Hat Cluster Manager and LVM for Cyrus IMAP
Sebastian Hagedorn <Hagedorn <at> uni-koeln.de>
2002-08-27 15:09:48 GMT
we're currently planning a big scale migration of our mail system. At the
moment we use two different setups for students and for staff, both running
sendmail and UW IMAP and POP.
We want to move to one system using sendmail and Cyrus 2.1.x. There will be
about 50,000 users. The current plan is to use a cluster of altogether four
Dell servers running Red Hat 7.3. The MTA-Out and some other services will
be load balanced on two of the machines using LVS aka Piranha.
The MTA-In / IMAP / POP part will not be load-balanced, but the idea is
rather to increase its availability. This we plan to achieve by using the
Red Hat Cluster Manager as Failover Service. The two servers will be
physically connected to an external RAID box with two SCSI controllers.
When one of them goes down the other one mounts the required volumes off
the RAID box and starts the configured services.
Has anybody on the list ever done something like this?
Another question I have regards the partitions for the Cyrus mailboxes.
Using the Linux logical volume manager (LVM) and XFS's xfs_grow command it
is possible to grow partitions while the machine is running and they are
mounted. To me it seems easiest to use just one such partition for the
mailboxes. Because the *physical* arrangement of the data can be adjusted
at will using LVM this is very flexible. Are there any downsides to this
approach that I'm overlooking? The bottleneck will be the UW-160 SCSI
controller, but I would think that that would be enough ...
All comments are welcome.