"ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU> wrote on 06/20/2006 
03:05:47 PM:

> John Monahan a écrit :
> 
> >>Did you do any FILE devclass work?  It sounds as though your miserable
> >>performance was head thrashing, which we'd get in multiple LV DISK
> >>devclasses, and also in FILE devclasses.  Darn. :)
> >> 
> >>
> >
> >No I was using a DISK devclass, but I'm almost positive it wouldn't 
have
> >made any difference with FILE devclasses.  Some of the LUNs were faster
> >than others (first ones in the RAID group created) and as soon as I was
> >writing to more than one at a time my performance would tank even 
further,
> >which shouldn't be any different had I used FILE devclasses.  I would 
have
> >tried file devclasses with more time for performance reasons, but I had
> >spent too much time on it already and the 150 MB/sec writes for my
> >diskpool was adequate for my needs and outpaced the capabilities of my
> >single Gigabit network connection to the TSM server, so I moved on.
> >
> >Hundreds of LVs with FILE devclasses on fibre disks would probably do
> >exactly what you are looking for and provide excellent performance. 
SATA
> >just isn't good enough at multiple, simultaneous I/Os to the same set 
of
> >disks yet.
> >
> >
> > 
> >
> I remember that it's some limitation with DISK devclass for restore 
> purpose. It was a good IBM presentation about disk oriented backup with 
> good advice but I don't succeed to find the URL (TapelessTSM).

All the restores will be from tape.  This is a traditional diskpool, that 
gets emptied out to tape each night so the data doesn't stay on disk for 
more than a few hours.  This particular customer has over 1TB to backup 
nightly, and not that much fibre space available, so SATA it is.