We're considering using ADSM's HSM to make 1.4TB of data
accessible for our users.  Our users do their work on a
SparcServer 1000E, which has ~200BG of disk.  Our first
thought was to install ADSM and a tape library on the SS1000E,
add Space Management to one of the filesystems, load that
filesystem with data, let ADSM migrate the data to tape
and then the data would be recalled as users reference it.
Unfortunately, ADSM HSM currently requires a single processor
machine and the SS1000E is a multiprocessor.

Now we're considering having the SS1000E NFS mount space
from a single processor SparcServer 5.  On the SS5, we'd
install ADSM and a tape library on the SS5, add Space Management
to one of the SS5's filesystems, and load that filesystem with data,
which ADSM migrates to the tape library.  The SS1000E would
NFS mount that filesystem.  Users would still do their
work on the SS1000E.  I've looked through the ADSM-L archives,
but couldn't find much discussion of NFS mounting a Space
Managed filesystem.  My concern is how well NFS interacts
with the Space Management.  Does migration cause NFS Stale
Filehandles?  Will NFS timeout waiting for recalls?  (The
data files range in size from 200MB to 1.5GB, so a recall
from tape can take many minutes.)  If a user ^C out of
an application which is waiting on a read which is waiting
for a file to be recalled, will the application exit or
hang on the read?

We're experimenting with the ADSM try-buy, but maybe someone
already has been through this and can share their results,
or give me hint where to look in the ADSM-L archives for
previous discussions.

Regarding why ADSM HSM requires a single processor machine,
is this simply because IBM hasn't yet ported the Space Management
filesystem driver to be MT-safe, or because of some intrinsic
limitation of ADSM Space Management which will prevent it
from ever running on a multiprocessor?

Thanks,

Ron Stanonik
stanonik AT nprdc.navy DOT mil