I've done (or seen done) a couple of different things to minimize this in
the past.

1.  Segregate data in copy pools.  If you have TDP backups and flat file
backups going to the same pool, your TDP restore may have to wait for a
tape that a flat file restore is using.  Having different primary and copy
pools for these types of data fixes that issue.  You can then also
collocate the copypool for the TDP data, virtually eliminating any
conflicts there.  You can even do the same thing with groups of clients, if
the need presents itself.
2.  If you're doing any type of periodic "full backups" on the clients,
stagger them on different days.  ("Full backups" being archives/selectives
to give a restore baseline, then restoring from the incremental to catch
things up).  This way you aren't hitting the same tape for a number of
clients all at once.
3.  Don't run more concurrent restores than you have tape drives.  You can
only restore as fast as your tape drives can read the data (or as fast as
your network will allow).  It won't actually go faster (and could be slower
if you have gaps between restores ending and starting), but you'll see less
wait time on the clients, so it will seem faster.  And that, when your
customer is watching the progress, can be very important.  You also finish
individual machines along the timeline, so you can give progress reports
that have facts in them, which also makes the customer happy ("We've
finished restores on 20% of the machines," sounds much better than, "We're
about 20% done with the restores.").

Nick Cassimatis
nickpc AT us.ibm DOT com

Think twice, type once.




                      Ted Byrne
                      <[email protected]        To:       ADSM-L AT VM.MARIST 
DOT EDU
                      OM>                      cc:
                      Sent by: "ADSM:          Subject:  Minimizing tape 
contention at DR
                      Dist Stor
                      Manager"
                      <[email protected]
                      .EDU>


                      05/20/2003 11:10
                      AM
                      Please respond to
                      "ADSM: Dist Stor
                      Manager"






Greetings,

We have been running into restore performance problems at DR, with multiple
clients waiting on the same tape cartridge, and an excessive number of tape
mounts.

We are attempting to minimize the issue with the number of tape mounts by
aggressively reclaiming the offsite storage pools, and have implemented the
SEARCHMPQUEUE option on the server to take advantage of that functionality.

We are considering an additional possibility, which is to make use of the
second copy storagepool by loading those tapes into the library along with
the first copy storagepool.  The question that we have is whether TSM will
utilize an additional copy of an object to satisfy a restore request if the
first copy is on a volume that is "in use" rather than offsite, unavailable
or destroyed.

For example:
        Node1 has files being restored from tape ABC (in stgpool VAULT_1)
        Node2 also has files on tape ABC.
                There are other copies of these objects
                on tape XYZ (in stgpool VAULT_2)

        While tape ABC is in use for the restore of Node1's data, will
        TSM mount tape XYZ to satisfy the restore of Node2's data?

Thanks in advance,

Ted