Zoltan:
Have you looked at the server-side of this situation? You can optimize
for backup or restore but not usually both. While you may slam TBs to
the server in a short window of time, you usually are increasing the
window on the restore-- unless you've thought about it when
architecting.
Do you have the restore waiting for a bunch of mount points?
Would it work to try to use the TSM server to create backup sets for
"chunks" of the client's data? (Never tried this, but it always sounded
like an intriguing concept with the backup set idea.) Or to attempt
parallel restores on data that doesn't reside on the same tape?
Have you looked to see if collocation on the stgpool is an issue? If you
are "single-streaming" from one tape, can you connive a way to set up a
diskpool with that clients' data to "prestage" the restore? Do a copy
stgpool (if on tape) into a bunch of disks that you can "spray" the data
back onto the client?
If none of this is helpful, I apologize and I wish you the best of luck!
I am on the client side of things TSM now, so I may not have a clue
anymore! (or any less...)
thanks!
lisa
> -----Original Message-----
> From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf
> Of Zoltan Forray/AC/VCU
> Sent: Thursday, December 10, 2009 1:10 PM
> To: ADSM-L AT VM.MARIST DOT EDU
> Subject: Re: [ADSM-L] Restoring LARGE server
>
> AFAIK, NQR isn't something you can "pick and choose" - if it was, I
> would.
>
> From the book:
>
> When you enter an unrestricted wildcard source file specification on
> the
> restore command and do not specify any of the options: inactive,
> latest,
> pick, fromdate, todate, or |volinformation, the client uses a no query
> restore method for restoring files and directories from the server.
> This
> method is called no query restore because instead of querying the
> server
> for each object to be restored, a single restore request is sent to
the
> server. In this case, the server returns the files and directories to
> the
> client without further action by the client. The client merely accepts
> the
> data coming from the server and restores it to the destination named
on
> the restore command.
>
> What is killing me is having to restore inactive file (the server
> corruption completely killed a big directory that the OS says is now
> "empty" so the TSM backup marked everything "inactive").
>
> In my opinion, these restrictions for NQR plus the issue of not having
> an
> equivalent of MEMORYEFFICIENT for restores, is a major
problem/weakness
> in
> TSM.
>
> I have spent the past 3+ days (and will be working throughout the
> weekend)
> to restore these files. I keep having to drill further and further
> into
> each directory structure to guess which directory I can attempt to
> restore. Every time I click on a directory and attempt a restore, the
> process runs for 30-minutes transferring 2G or more of meta-data
before
> it
> eventually fails on "insufficient memory" or I am lucky enough to pick
> one
> that restores. Then I have to start all over again, keep track of
> where I
> am and what I was able to restore, drill further down,
> rinse-lather-repeat! Besides the 22M objects, there is 700GB of data
> to
> process/wander through.
>
> What good is having 50M objects of backups if I can't restore them in
a
> reasonable fashion?
>
>
>
> From:
> "Laughlin, Lisa" <Lisa.Laughlin AT OA.MO DOT GOV>
> To:
> ADSM-L AT VM.MARIST DOT EDU
> Date:
> 12/09/2009 11:15 AM
> Subject:
> Re: [ADSM-L] Restoring LARGE server
> Sent by:
> "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
>
>
>
> How about using just the CLI and the no-query restore option?
>
> thanks!
> lisa
>
>
> > -----Original Message-----
> > From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On
Behalf
> > Of Ochs, Duane
> > Sent: Wednesday, December 09, 2009 9:57 AM
> > To: ADSM-L AT VM.MARIST DOT EDU
> > Subject: Re: [ADSM-L] Restoring LARGE server
> >
> > Zoltan,
> > Have you attempted a Point in Time restore from command line?
> > That might help with the number of inactive files you are
> experiencing.
> > If that is not an option, you may have to go a couple directories at
> a
> > time.
> >
> > I have only had experience restoring up to 9M files and the one time
> I
> > did it used PIT and worked fine, turned maxmp up to 4. It certainly
> > took a while. But it did complete.
> >
> > What OS are you using on the client ?
> >
> >
> > Duane
> >
> >
> >
> >
> > -----Original Message-----
> > From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On
Behalf
> > Of Zoltan Forray/AC/VCU
> > Sent: Wednesday, December 09, 2009 7:51 AM
> > To: ADSM-L AT VM.MARIST DOT EDU
> > Subject: Restoring LARGE server
> >
> > Trying to restore a LARGE Windows server. Over 40M objects. Client
> is
> > 6.1
> >
> > As you can imagine, we have had to use the journal as well as
> > MEMORYEFFICIENT to perform backups.
> >
> > If I read correctly, MEMORYEFFICIENT is ONLY for backups. Obviously
> > the
> > journal is of no value since the restore is to a new
> machine/location.
> >
> > The other issue gumming up the works is that the backups have been
> > failing
> > (drive array problems/corruption) and thus TSM has marked almost
> > everything
> > as "inactive". A first, NQR restore (just selected the drive)
> attempt
> > only
> > restored 90GB (of 600G+) before "finishing successfully". But when
I
> > choose to pick inactive as well as active, the NQR is disabled.
> >
> > The server transfers down 1GB of metadata before the client chokes
on
> > "insufficient memory".
> >
> > So, how am I supposed to restore this many objects, besides picking
> one
> > directory at a time, which is not to say that some directories are
> > large
> > enough to also cause a "not enough memory" situation?
> >
> > Suggestions on how to handle this?
|
