Looking for advice on migration to new Cyrus-IMAPd server
Scott Lambert
lambert at lambertfam.org
Fri Mar 11 18:54:14 EST 2011
I've built a new Cyrus-IMAPd server (hardware). It has two sets
of mirrored disks. The OS is on WD VelociRaptor 450GB drives. The
bulk IMAP data will be on WD Black RE-4 2TB drives. So far, I think
I want to use ZFS for both mirrors.
I wanted to see what the list recommends for the migration.
The current server is running FreeBSD 8.2, with cyrus-imapd-2.3.16
on a RAID-5 setup on SCSI-UW with 15000RPM drives. We get about
1.5 - 9MB/s at 80 - 100% busy according to "systat -vm 1" for most
of the day.
The existing server has no metadata partition. On the new server,
I believe that I want to put the config directory and metadata
partition on the 46050GB drives and the message files on the 2TB drives.
Can I configure the new server's imapd.conf to put data where I
want it, and simply use the cyrus replication to put the data in
the right place on the new server? In other words can the migration
split my data partition into data and metadata partitions without
my having to script up shuffling files around then reconfiguring
and restarting cyrus?
Also, should I install Cyrus-IMAPd 2.3 or 2.4 on the new server?
I tend to be a late adopter. I'm wondering if 2.4 is generally
considered ready for a single server install in a multi-domain ISP
type environment.
I know Bron mentioned that fastmail is using 2.4, which lends a
warm fuzzy feeling to it. But they have the multiple brains on
staff to identify and resolve problems quickly. I have about
1/1000th of a brain to dedicate to mail server problems once I move
on to the other projects which are in the queue. It would be nice
to be able to fire and forget. I haven't had any problems with
2.3, just want to be on 2.4 for locking improvements.
It sounds like 2.4 would let my users configure their iPhones for
POP3 without creating tech support calls when the constantly attached
iPhone locks out their desktop computers from checking mail. That's
my most desired feature after splitting the data and metadata to
separate spindle sets.
--
Scott Lambert KC5MLE Unix SysAdmin
lambert at lambertfam.org
More information about the Info-cyrus
mailing list