>Veritas recently commishioned a "study" of performance between
>Netbackup, Networker, and TSM to compare results of snapshot backups.
>Apparently the new Netbackup 5.0 has a new "advanced client".
>For TSM, they threw it in stating that there is no comprable feature,
>but they wanted include it anyway.  I have never done any investigating
>into doing snapshot backups of data with TSM, do any of you do anything
>similar at your site and if so, what?
>
>Link to the marketing trash:
>http://veritest.com/clients/reports/veritas/veritas_backup_w_add.pdf
>
>Michael French
>Savvis Communications

Good morning, Michael -

Many of us don't much look at competing vendor offerings, so it was interesting
to learn of the new Veritas FlashBackup approach.  I read through the test
report, and searched out the basal information at
http://www.veritas.com/van/products/nbuflashbackup.jsp
to get more information.  Some observations...

The VeriTest report is troublingly vague on just what Veritas contracted them to
test.  The impression is left that VeriTest themselves determined just what
would constitute the test environment.  We are given no information as to what
their "large data set" is supposed to represent (active web server HTML and
programs file system; user home directory file system; mixed database
environment; mail spools?).  Reading all the material, however, it is apparent
that the test is to show the advantages of only FlashBackup within the narrow
focus of backing up a large number of modest files (representative of user home
directories, with their frequent file changes).

One thing that jumped out at me was that the test chose to use DLT tape
technology.  While DLT or LTO may be a relatively common deployment choice, it's
well known that as principally streaming media that their performance drawbacks
will prolong the backup of more than a million small files.  This will
inherently make any conventional file-oriented backup approach look bad in the
test.

Almost comically absent from the test was the real reason that one performs
backups: restoral.  We've all seen vendor literature touting one aspect of a
product without presenting the overall whole view.  The FlashBackup whitepaper
(unreferenced in the testing document) suggests that restoral time is comparable
to the product's historical norm, but certainly one would like to see that
included in such a test.

I laud the innovative approach that Veritas has taken with FlashBackup: it's
always good to see imagineering occurring in product development.  I'm concerned
about the durability of this approach to file backups, though.  FlashBackup
operates by first capturing a file system map (it has to run through the file
system like other conventional products, and so can get somewhat bogged down in
directories).  Then it separately writes disk blocks to tape, bypassing file
system I/O overhead, whence it gains its speed.  This amounts to a kind of file
system emulation.  Is this foolproof, particularly in highly volatile file
systems?  Will this work under all circumstances and all kinds of file system
objects?  This is what we would have liked substantive, thorough testing to tell
us.  We aren't getting that information from the vendor or their contracted
testing company, so it will apparently have to be proved out in the field.

Some of the information surrounding FlashBackup can mislead.  The testing paper
states that IBM's TSM "...does not have an online snapshot capability..."
Strictly speaking, that is true.  TSM does, however, provide image copy backup
and restoral.  And there is the Tivoli Storage Manager for Hardware product,
which takes advantage of the IBM ESS's Copy Services to facilitate DB2 and
Oracle backups.  This leads to the question of whether you want "flash" type
backup capability in your backup product, or native to the file system
infrastructure.  Some file systems, such as AFS and DFS, are specifically
architected to support backups through a flash mechanism.  More hardware is now
accommodating this need, particularly in 24x7 operations.

I would strongly advised reading FlashBackup's whitepaper, which gives a much
more balanced sense of where FlashBackup offers advantages, and where it does
not.  In particular, note that, in contrast to more generalized backup products,
FlashBackup works only with certain file system types.  It would be most helpful
to get some feedback from its beta testers.

  Richard Sims