ADSM-L

Re: TSM subfile backup vs TSM incremental ?

2003-09-09 11:05:05
Subject: Re: TSM subfile backup vs TSM incremental ?
From: "Prather, Wanda" <Wanda.Prather AT JHUAPL DOT EDU>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Tue, 9 Sep 2003 11:04:36 -0400
>>Question: does anyone have "practical" experience with subfile backups?

Yep - about 350 Win2K machines using it for last 2 years.
Our server is still 4.2.1.15, clients at 4.2.0.0; haven't tried it yet at
5.1 server.

>>Is it easy to implement?

Yep.  Really straightforward.
Easy to test, easy to implement.
Don't forget you have to SET SUBFILE ON at the server end.
We use subfile for all files on the machine.
But, you can restrict it to certain files using include/exclude.subfile
options.
We started out using it to attack the problem of large .pst files on many
machines.
Cut our overall nightly load by about 30%, probably extended the life of my
server by a year.

>>Is there anything to pay special attention too?
*       it works by creating a /cache directory under the /baclient
directory; MAKE SURE you exclude that from backups!
*       DON"T FORGET that subfile DOESN"T HELP YOU on restores!  Make sure
you manage your user expectiations in that area.
*       If you aren't using collocation on the server end, better think
twice - for every restore, you have to restore the base and the last delta
file.  Will lead to extra tape mounts without collocation.  We found it
didn't cause any noticeable increase in restore times, but we use
collocation and very fast tape.
*       you probably saw that it doesn't apply to files over 2 GB, just be
aware of that.
*       we also had occasional problems where a machine would stop backing
up because the subfile cache (on the client end) would get corrupted.  But
the easy fix was just to delete the entire cache directory and let it build
a new one on the next backup; happened so infrequently we never tried to
pursue a fix with IBM.
*       the stats in dsmsched.log show the size of the ORIGINAL file, not
the size of the subfile that actually got backed up.
*       Only the backup-complete stats will ever give you an idea of how
much data actually got sent.

I whacked it around, rebooted in the middle of a backup, deleted the cache,
did everything I could think of to break it.
NEVER could cause errors on restores.
There were problems in 4.1 and 4.2 servers where the subfiles didn't expire
completely when they should, but never any data integrity problems.  We do a
lot of full restores, right down to the user profile and the last icon, and
never had a problem.

You'll like it!