==> On Mon, 5 Dec 2005 16:33:43 -0500, "Spearman, Wayne" <wmspearman AT 
NOVANTHEALTH DOT ORG> said:


> We are planning to move a large number of our nodes to "disk only" backups
> and eliminate tape for them. The diskpools will be devtype=file.  I struggle
> with whether to collocate these disk volumes or not and how large to make
> them. I've had some discussions with IBM, but would like to ask others to
> share their experiences with me.


For collocation, my question would be 'Why ever not?'  If you've got N
machines backing up simultaneously, you'll have N files open; I'd say pick the
"correct" N.

As for size, My aesthetic opinion on this is that they should be substantially
smaller than 1% of the enclosing capacity.  So if you've got a 1TB of space, I
think a 10GB volume is too big.

The last time I actually picked a number, we had ~8TB of space, and I picked a
5GB volume size.  This meant that my reclamation was happening in what felt
like manageable chunks, and if I wanted to actually move things around I could
do so on a sane granularity.

The other force on volume size is "I don't want to have too many to manage".
I'll assert that once you get into disk-only backups, you're going to have a
volume count which you will -need- to manage in an automated fashion.  So you
_will_ be writing scripts to do things, instead of typing e.g. "move data"
commands at the TSM> prompt.  SO since you've already lost the
manual-operation battle, you may as well go for the "reasonaby small" win.


Keep in the back of your mind that backupsets are portable; if you write a
backupset to disk, and then copy it somewhere else, you can use it
indefinitely. So the FILE storage might be accessed from somewhere other than
thru TSM.



- Allen S. Rout