We've got a multi-terabyte (20+) Oracle database that we're trying to backup
via RMAN.  The host systems are hammered so we get frequent failures due to
resource issues.  It also takes the better part of two days to do a full
backup (Across four direct-access LTO-1 drives).  

Another problem we're having is redo-log accumulation during backup.
Two-day backups make too many archived redo logs.

We're looking at some changes in backup strategy so I'm looking for some
advice.

Restartability.  Right now, our full backup script simply calls the RMAN
command: 

   RUN { BACKUP INCREMENTAL LEVEL=0 SKIP INACCESSIBLE 
         TAG '_full' FILESPERSET 8 FORMAT '%d_%s_%p_%t' 
         DATABASE PLUS ARCHIVELOG DELETE INPUT;}

Things we're look at and talking about:

I need a way to identify when a previous full backup attempt has failed and
initiate a new backup of datafiles that were not yet backed up.

Breaking into sets.  Is it practical to break the database into sets of
tablespaces and spread the full backups over a week with differential
backups for those tablesspaces that the nightly full misses?

Breaking backup into per-tablespace backups with archived-redo handled
between tablespaces.  This would solve the accumulated archived-redo problem
but seems a management nightmare.  Also, doing 8 streams like we do today
also seems much more difficult with this strategy.  Perhaps, breaking into
simultaneous sets of 8 tablespaces with redo-log backup in between?

Finally, what's a good RMAN book for non-DBA's?  Looks like I get the RMAN
piece as the backup admin even if I'm not a true DBA.

-M

----------------------------------------------------------------
Although I can accept talking scarecrows, lions and great wizards of
emerald cities, I find it hard to believe there is no paperwork
involved when your house lands on a witch.  --Dave James
------------------------------------------------------------------- 
Mark Donaldson - SE - Experian
-------------------------------------------------------------------