ADSM-L

Re: Stgpool backup behavior change in 5.3.x?

2005-12-28 21:22:39
Subject: Re: Stgpool backup behavior change in 5.3.x?
From: "Allen S. Rout" <asr AT UFL DOT EDU>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Wed, 28 Dec 2005 21:22:36 -0500
>> On Wed, 28 Dec 2005 13:49:47 -0500, "Thach, Kevin G" <KThach AT COVHLTH DOT 
>> COM> said:

> Hi Kevin, I have finished my review and, per our phone conversation, I
> believe I know exactly what is causing this performance problem. The fix for
> APAR IC45931 was introduced in the 5.2.6.2 level of TSM Server code (which
> you are running) and part of this fix is to prevent our backup optimization
> processing from invalidly skipping files on a volume which have not been
> backed up. This is causing a performance problem in your environment since
> we have to re-examine alot of the same files during each backup stgpool
> run. This examination is done under a thread which is not registered as a
> process which is why you see what appears to be a hang after you issue the
> backup command. The extent of the performance degradation is dependent on
> how many volumes are in the storage pool and how many do not have current
> optimization entries. In your case, I would say 80% of the volumes being
> analyzed are not current.

I understand the re-examination optimization; it actually makes very good
sense.  Seeing it cause pain in this circumstance makes me rethink the target
size of my collocation groups to avoid near-100% "not-current" volumes.  But
that's a different topic.  :)

But that is about "Why is backup stgpool taking a long time and not doing any
work", which behavior I feel to be not a bug.

"Why isn't my 'background' command returning after 8 hours" is a different
question and, I claim, a bug.

"Why isn't a process which runs for 8 hours and obstructs certain other work
in the process table" is another different question.

I think that if the process runs for more than a few seconds, _certainly_ a
minute, it should toss itself into the background and render itself up for
management like every other process.

Better yet, the bloody thing should 'WAIT=NO' if it's set to wait=no, and then
they could offer 'WAIT=only_as_long_as_I_have_no_work' as an additional
value-added option.  Hmph.


- Allen S. Rout
- Markteting can rename that option, just like MEMORYEFFICIENTBACKUP.

<Prev in Thread] Current Thread [Next in Thread>