On Wed 11 Oct, 2000, Marion Hakanson hakanson@cse.ogi.edu wrote:

I think things are headed toward the use of SAN-connected tape drives and libraries for this type of setup. The backup software tells the SAN switch to connect up the appropriate number of drives to the backup client (NetApp), and then uses NDMP to initiate the backup on the client, using "direct attached" tape drives. When that backup is done, you "direct attach" the drives to the next client.

I can think of a rather neat scenario along those lines: as described tape drives and disk drives are all on a switched FC fabric along with one or more NetApp filers, and possibly an NDMP backup host or two.

When a backup host starts a backup running, the NetApp starts causing blocks to be transferred directly from the disks it manages to the tape drive(s) that the backup host controls, mingled with a little metadata it originates itself, while sending the lion's share of metadata to the NDMP backup host.

Nothing should go across any piece of wire any more times than is absolutely necessary.

The same scenario could go for any normal open system (Unix, NT, AS400, whatever) in a similar fabric, but I'm not sure who's best placed to actually make it work. I'd like to think NetApp were.

See, until these scenarios start coming true, SAN's aren't much more than an overworked buzzword for big flipping messes made of disks and fibre. I still think SAN's are an about-to-be-useful item, but I've yet to see much real progress.

Heck, some libraries (ATL, for one) are becoming available with NDMP servers right in them, ethernet interface included. Dunno where you put the file history or tape indexes, but it's getting interesting.

With the nominated backup controller boxes, which keep the metadata on-line (on disk) so that restores, archival, cloning etc. remain first-order activities, as much as running databases and maintaining other applications.

Backups should never be afterthoughts as they so often are. Sorry, got a bit bitter and twisted there. 8)

Speaking of NAS & tape libraries, does anybody have any experience with the Grau Data "Infinistore Virtual Disk System" (www.graudata.com)? It's apparently a NAS box with a RAID cache in front of a tape library.

Sounds like fun. Crafted well, the scenario I describe would ideally have the IO balanced to keep the tape drives well-fed, the backup window small, the filer heads serving out of the front-door performantly and the backup controllers adequately busy. Otherwise, what would be the point? Strikes me that they're angling toward HSM which is related but not the same.

The brief article I saw in Server/Workstation Expert magazine mentioned that an "entry level" system starts at around $100k, so I didn't read beyond that (:-). We sure wish NetApp had something like this, since somewhere around 80% of our filer's capacity is tied up in files that haven't been accessed in a year or more.

And that final comment makes me think you'd quite like an HSM product with NetApp-style qualities. I've often thought that it'd be nice if NetApp could make such boxes, by way of extension beyond the filers, but I don't know if anyone's tested the market to see what sort of demand there might be.

-- End of excerpt from Marion Hakanson