Env: DB2 9.1/AIX 5.3
Database size: 750 GB
No of tablespaces : 40

Complete offline backup runs for 3 hours with the following options
UTIL_HEAP_SZ = 5000
backup db xxx use tsm open 2 sessions.

Which is a best option to improve the performance
[1] Increase UTIL_HEAP_SZ to 10,000 ?
[2] Use more sessions with TSM (4) ?
[3] WITH num_buffers BUFFERS parameter ?

Any thoughts ?

You can find some general performance guidelines here:
DB2 Database for Linux, UNIX, and Windows


OPEN n SESSIONS
Limited by maxnummp TSM parameter so # of sessions should not be greater than maxnummp


WITH num-buff BUFFERS
The following calculation must fit: (num-buffers * buffer-size) < util_heap_sz

Or you can let DB2 choose the value, if it's not specified. You can see what db2 used with diaglevel 4.


You can also consider using incremental/delta backups. A tablespace will be skipped if there are no changed pages.

The problem is usually in the network between db and TSM plus the inherent inefficiencies of the TSM API. Backup to disk (small fraction of API time) and have TSM pick it up from there. Of course, with this scenario, you have to be prepared to put files back on disk from TSM manually before doing RESTORE.