Re: [ADSM-L] Objects Assigned vs. Your Database.

2012-04-16 23:30:44
Subject: Re: [ADSM-L] Objects Assigned vs. Your Database.
From: "Prather, Wanda" <wPrather AT ICFI DOT COM>
Date: Mon, 16 Apr 2012 21:08:00 +0000
Another bandaid:

For Win2K8 clients backing up to 5.5 servers, you can use the 
SYSTEMSTATEBACKUPMETHOD parm described below in dsm.opt, to essentially force 
the systemstate backup to work like it used to (all systemstate all of the 
time), or "opportunistic", (all systemstate if anything has changed, no 
systemstate if nothing has changed), instead of doing a true incremental of 
systemstate and thrashing your TSM data base to death.

But from experience at a customer where we had similar problems (Win2K8 clients 
were taking 8 hours for the incremental systemstate backup), the long-term 
solution is to get your clients to a V6.2+ TSM server.  Problem goes away, 
incremental of systemstate works fast, no need to specify 

-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf Of 
Rick Adamson
Sent: Monday, April 16, 2012 3:32 PM
Subject: Re: [ADSM-L] Objects Assigned vs. Your Database.

I had the same issue on my 5.5. servers backing up 6.2.x.x client system states 
and was able to remedy the situation by moving my database and log files to 
faster drives.

Historically, my TSM servers have not had sufficient local storage to 
accommodate these volumes (as well as cached disk storage) so they lived on an 
EMC Clariion, which had both SATA and Fibre drives. At some point during a San 
upgrade it appears that the TSM storage was moved to the slower SATA drives.
I had been chasing pinning logs and SLOW system state backups for months and 
last week as a test we decided to move the TSM database, logs, and disk pools 
to an EMC Vmax. Since then log usage has not been over 20% and system state 
backups are completing with almost no lag time.
Comparatively, for 3-4 months the log files have been at 85-98 percent 
utilization every night, which as you probably know usually end up with tasks 
being canceled to prevent a total system failure.

While I realize that everyone may not have the option of faster storage 
available I thought it important to re-emphasize how the storage layout and 
disk speed for the TSM database/log storage can affect server operation.

Before the test/change system state backups would appear to hang for several 
hours while calculating what needs to be backed up, and any large files would 
pin the log.

Jacksonville, FL.

-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf Of 
Allen S. Rout
Sent: Monday, April 16, 2012 2:22 PM
Subject: [ADSM-L] Objects Assigned vs. Your Database.

Howdy, TSM folks.

So I think I've gotten to the bottom of a performance issue I've been seeing 
recently (and a crash!) and I wanted to compare notes.

Cutting to the chase, I've been seeing obnoxious log consumption on one of my 
TSM servers every night recently, and once a few months ago it got bad enough 
that it blew out the log and crashed with the 'gotta extend
the log' situation.   Routine recovery, but irritating.

At the moment, I'm attributing the problem to a growing population of v6 
clients which, when doing system state backups, are "reassigning"
objects from the previous backup to the current one.

Now, I understand why they're doing this:  It gets us back into the incremental 
zone, from the miserable 'system state :== full' situation.
    But the law of unintended consequences comes in stage left.

Since many (most?) of the system state is in fact static, that means that each 
machine is going to reassign most of its system state.

How fast can it do that?  How fast is your database?  That's how fast.

Last night I watched what felt like a normally busy evening, and the log-full 
percentage was growing before my eyes;  as in, wait a minute and see three 
percent advance, and that's on an 11692MB log.

So I've got my trigger set at 60%, but it's blowing through the remaining 40 
like nothing.  I get to the 'server log is [foo] full, delaying transactions' 
state on a regular basis.

As a band-aid, I'm going to talk to the customers and see if some of this 
population of machines can rationally be excluded from SYSTEM STATE
backups:  Offhand, I think their DR plans don't include BMR from my TSM.
  But that's short-term thinking.

What are you-all doing about this?  Increasing number of DB incrs between 
fulls?  Something else?

- Allen S. Rout

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