Hello *SMers,
An IBM business partner said to me that restores are what TSM is plugged for.
My feeling is that on low number of file restores I can understand, but on
restores for large number of files there is much to be desired.
I can not find any one post on the listserv that sums up how to optimize
restores. If you have any stored messages please let me know. Anyhow, to beat
a dead horse with restore performance issue, I wanted to brainstorm some common
practices not specific to hardware. If there are common hardware settings not
specific to any tape drive please say so. The following is that brainstorm,
please fill free to add, dispel any statements, or general thoughts.
Collocation - Generally used for large files stores or critical servers. Too
many large nodes using a collocated pool would defeat its purpose to keep node
data on as few of tapes as possible. The more collocated pools, the more
backup, reclamation, migration, etc... processes involved on the TSM server.
DIRMC - Good for large directory structures. Should it be used on all servers
large and small? Should the dirmc management class keep as many versions as
the longest management class containing data so that an old file can be
restored with its entire directory structure? Are most organizations using
this on ALL servers?
Disabling Antivirus - Depends on individual anti-virus configuration, but
disabling write scanning during restores is always good for restore performance.
Command-line Classic restore - Have yet to use on large restore, haven't seen
any performance spec's on CmdLine vs. GUI. Certainly can understand how it
could be faster if commands don't have to go through a GUI front-end.
Move nodedata - I'm currently at 4.2.1.9(IBM NSM) so I haven't had the
opportunity to use it. I believe the theory behind it is to defrag nodedata so
that is on as few of tapes as possible. It says you can do this within the
same storage pool but I am curious to hear about anyone who has used it. I
could see this as a good process to run on a weekly or monthly basis depending
on the size of the shop.
Lowering Mount Retention - This definitely helps so that the tape is held as
short as possible by shortening the time between mounts of tapes during
multiple mount restores.
Optimizing resourceutilization - I am not sure if there is an actual formula
for optimizing. ( i.e. (1 process/tape drive), etc...)
Multiple Restores at a time - Is there a formula for this (i.e. 1 restore/tape
drive)? Theory makes sense for breaking up large restores.
HSM products - Makes sense to have 40-60% of data already on tape and not have
to restore in a disaster. Currently we are trying to implement DX2000(I know
Legato owns it, politics have pushed this solution), but backup and overall
performance of the box has crawled while drivers and services are enabled.
Thanks in advance for any feedback,
Bud Brown
Denver Health
Information Services
Systems Administrator
|