> > When you mirror the db and log, use ADSM mirroring.  Also, on your last
> > question, are you asking about using volume manager to glue the disks
> > together?  If so, don't.  Just create an ADSM stgpool volume per disk.
>
> It depends. We decided to have the stage on two RAID-5 sets for
> different policy domains on an external HW-Raid. Otherwise you might
> loose the unique copy of a file by single disk failure when there is no
> redundancy on the backup server side. But everything here is one order
> of magnitude smaller, except the computer ;-) So maybe RAID-5 is to
> time consuming or maybe you can guarantee that you have two copies
> of the data at any time. In that case disk striping could be faster than
> having one volume per disk, depending on the number of sessions and
> migration processes. I am not sure wether *SM makes any attempt to
> distribute sessions or migration processes to disk volumes.

Raid-5 has a write penalty that will affect performance.  If you do not
mirror/raid the disk storage pool, sure, until you backup that storage
pool, the *SM server only has one copy, but the client machine should
still have a copy.

Striping is scarry, if you ask me.  *SM does it's own "stripping" of files
across the volumes in the disk storage pool.  If you create a strip in the
hardware, and do not have redunancy in that hardware, what happens if you
loose a volume in the strip?  *SM's view of it might actually cause it to
have errors on all of it's disk volumes, meaning you would have to audit
every one of your storage pool volumes (instead of only one), or even
loose all of the *SM volumes.   Auditing the disk storage is not a fun
task, and time consuming, especially if caching is enabled.

If stripping in hardware, I would recommend also mirroring that strip in
the hardware...

Steve Roder, University at Buffalo
VM Systems Programmer
UNIX Systems Administrator (Solaris and AIX)
ADSM Administrator
(tkssteve AT buffalo DOT edu | (716)645-3564 | 
http://ubvm.cc.buffalo.edu/~tkssteve)