Richard,

I'll try to get you some detailed information from our Cyrus support
person.  We have been running Cyrus for years.  A couple of years ago
we had some filesystem corruption and had to restore a mail spool
filesystem from TSM.  It took much longer than we wanted/expected it
to.  Yes, it's just the "restoring large quantities of small files" problem.

We have upgraded our TSM DB disks to FAStT 900, 15K rpm, and that has
helped some.  Parallel restore sessions helps some.  Using 'testflag
disable nqr' helps some.  Other things we have thought about, but
have not yet implemented, include:

- using vTape or serial access disk volumes, for keeping the email
backups.  Most of the performance penalty is involved in tape
skipping.  When restoring dis-contiguous files off of tape, it takes
time to reposition the tape.  This adds up.  This penalty is avoided
with vTape.

- using a combination of image backup and filesystem
backup.  Restoring an image first, then applying incremental
file-based backups should be faster than just restoring everything at
the file level.  Of course, if your corruption is backed up with the
image, you can still resort to the file-level backups.  The image
doesn't have to be in TSM, but could also be storage-system based
(i.e., flash copy).

..Paul

At 09:32 AM 5/16/2006, Richard Sims wrote:

We're a site who has been using IMAP with mbox style mail "folders"
for a decade. It's obvious that as we boost mail quotas, mbox "does
not scale". Thus, we are planning for a conversion to Cyrus, on Linux.

For planning purposes, I'd like to get input from Cyrus sites as to
issues they've encountered in performing backups and restorals,
particularly with coherency, given that each message becomes an
individual file. The Many Small Files issue is apparent, but there
are probably other things that sites have learned the hard way that
could help others. One unobvious issue I can think of is how to best
perform mail investigations via restorals, as called upon in
subpoenas. (A search of the ADSM-L archives turns up next to nothing
on Cyrus and TSM.)

I'd be happy to hear any other practical recommendations for
efficiently implementing Cyrus, based upon site experiences (disk
arrangement, file system type, directory architecture, etc.). Whereas
that would be off the TSM topic per se, feel free to email me
directly (rbs AT bu DOT edu).

   thanks, Richard Sims

--
Paul Zarnowski                            Ph: 607-255-4757
Manager, Storage Systems                  Fx: 607-255-8521
719 Rhodes Hall, Ithaca, NY 14853-3801    Em: psz1 AT cornell DOT edu