The only "sanctioned" method of backups requires either full db, else
transaction logs (diff or incr). The very best explanation of the choices
involved with Exchange are in the Install & User's Guide; also, see the
Redbooks for backing up databases with TSM -- and the many posts on this
forum from Del Hoobler.
Most dba's like daily FULL (online) backups -- regardless of the flavor;
that goes for Exch, SQL-server, Oracle (including SAP/R3 & Peoplesoft). The
larger the db gets, the more important it becomes to minimize the data loss
(to under 24-hours) between FULLs; mission-critical db's were on HA
machines, redo/transaction logs get dumped and copy-pooled every 3-4 hours
(depending on volume/time -- using filesystem monitor to watch for 30% of
capacity).
RMAN's not-so-great block-level incrementals have been slow on restore; the
incr speed is wonderful, getting good restore speed has been very
difficult/frustrating... if you really want block-level incrementals (for
ANY db, consider anybody's snapshot -- like Veritas, EMC-TimeFinder, etc.)
All things considered, you must evaluate the db-size and tape media involved
in context with your restore SLA; mostly, the dozen or so customers I've
had on Exchange go for daily or weekly fulls, plus (sometimes) incrementals
in between... but that was with Exch-5.5, smaller IS standards (always less
than 100 GB, strive to keep IS under 50GB, standard was 35 GB then put in
change control to create an additional server).
Now, with Exch-2K, the reliability & availability being much greater(?),
your IS will be best capped around 100GB, again depends on SLA for restore.
As your mileage will vary, per network, tape devices, and TSM server
capability, measure your environment for desired SLA --- on a shared LAN, we
readily achived 18 GB/Hr restore speeds (old 3575, 100 Mbps production LAN
during the business day at 3pm). Differential plus FULL versus several
incrementals after restoring FULL; BUT, wait... with Exc-2K, you don't need
to restore the entire IS, just the group that holds the mail objects you are
after ... so, given 80-95% restores are for mailbox of individual items,
you're passing more tape than actually restoring.
Logs are 5MB, number depends on how much activity you have; for my money's
worth, I'd advise (a) daily FULL (perhaps periodic INCR/DIFF during the day,
or (b) weekly FULL with daily DIFF (to consolidate logs for a given week --
in case you *must* do a restore, provided the number is "manageable" as
determined by your restore SLA!)
In our case, we typically saw less than 100 log files per day; for a week,
we'd accumulate maybe 500 or so. We used a separate storage pool, so each
day's data was separated only by other Exch. nodes on the same TSM server).
Good luck; you're gonna need to do your own homework -- there are no
shortcuts to due dilligence (on the customer's expectations and your
infracstructure to support any given SLA), especially for data base apps!
Don France
Technical Architect - Tivoli Certified Consultant
San Jose, Ca
(408) 257-3037
mailto:don_france AT att DOT net
(P.A.C.E. -- www.pacepros.com)
|
