On Mar 28, 2011, at 12:37 PM, andresms5 wrote:
> Richard,
>
> HSM is the architecture that I need. Do you know a place in the documentation
> of TSM where it states that restores are done in parallel when using HSM? My
> suspicion is that restores are being done in parallel, but they all converge
> in the server and after that they are copied from the server in parallel to
> the destinations. I'm I right?
Hi, Andres -
HSM does not involve "restores": its terminology is "recalls", where the whole
approach is different as each client action requiring a file constitutes a
separate request, as for example five people operating on their files in
different areas of an HSM-controlled file system. Recalls can be performed
intelligently and thus more efficiently where the client interacts with the
server to order a batch such that tape mounts (where needed) are minimized, as
can be achieved where the client is allowed to expand wildcards. And TSM will
first bring back files it knows to be in the HSM disk storage pool, to go after
tape-stored data after that. As the Unix HSM manual says: "When you
selectively recall a group of files, they are recalled in the most efficient,
time-saving order based on where they are stored."
Be aware that the convenience of HSM comes at a price: a great deal more
complexity, and system overhead, with a variety of service daemons operating
for and in the file system. Traversing the file system to find migration
candidates is particularly expensive.
Richard Sims
|