<<Question 1:
Interested in feedback regarding pros and cons VTL Vendors. >>
We have not yet gone into detail studying the (rather numerous) products,
but I have come up with a couple of key differentiators.
The ADIC Pathlight maintains a 1:1 relationship between virtual tapes and
physical tapes.
This gives a key benefit for DR as the tapes can be used directly by
NetBackup for restores. When NBU thinks a file is on tape ABC123, it is.
You can go to a DR site, re-create the backup server and restore from
tapes. The hardware requirement at the DR site is pretty flexible - a
suitable server and at least one tape drive of the right type - even a
single drive.
By contrast, VTLs that don't have this 1:1 relationship require that tapes
are first re-imported to the VTL, and can then be restored from there. So
DR requires that your DR site can host exactly the right brand of VTL, and
the restore process has two steps. So it has to take longer to recover
systems, and is significantly more complex.
-o-
There is of course a downside to the 1:1 relationship. As the Pathlight
supports multiple tape formats, and as the hardware compression
effectiveness varies across tape vendors (even within LTO I believe), it
assumes that data is not compressible. So a 200GB native LTO tape will
only map to a 200GB Virtual Tape. In reality the data may be 2:1
compressible, and if you had written it straight to tape you might get
400GB on that tape. So you could find that the amount of media you need
doubles ($$$), and of course we all hate the idea of half-empty tapes.
-o-
Other VTLs (especially with a mainframe background) were specifically
designed to improve tape utilisation. In particular where small
incremental backups simply did not fill a tape, so again mostly-empty
tapes are going offsite ( as most folk using tape will want last night's
backups offsite next day ). So these kind of VTLs write a tape when they
have enough on-disk to fill a tape, or maybe daily anyhow. As the VTL is
not trying to maintain any correspondence with the backup application
(indeed may be presenting partitioned robotics and so used by all sorts of
different software) it can send data to tape until it hits the EOT, taking
full advantage of the hardware compression.
Most tapes will be full, or at least as full as you allow. Media costs
are way down.
-o-
If you are rich you can have a pair of the latter kind of VTL, with
synchronous or asynchronous mirroring of the on-disk data to a remote
site. The tape copies are written on the remote site, where they can be
left in the tape silo (and it becomes worth having a big silo). As with
DR generally, you may have a single central DR site that has the tapes for
all your other sites, or you may have paired sites that back up each
other. You don't need to worry about providing the specific VTL on a
contracted DR site, as you are using your own DR site and VTL. So that
side is much simpler and quicker. But you do need a lot of bandwidth
between those sites, and that is why you need to be rich.
-o-
Like I said at the start, this is book theory and I'd hope others with
real experience of VTLs can chip in.
William D L Brown
|