ADSM-L

Re: DIRMC - Are copypool reclamation performance issues resolved or not.

2005-03-19 16:25:32
Subject: Re: DIRMC - Are copypool reclamation performance issues resolved or not.
From: TSM_User <tsm_user AT YAHOO DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Sat, 19 Mar 2005 13:25:11 -0800
Paul,
Using a separate pool for directories is something that many have been doing 
for a long time and just kept doing even after IBM implemented the new 
directory restore method (restore order processing).  If you look at a 
directory as a small file then you can see why keeping it in a separate pool so 
that you can keep that data on disk might help.  This is one reason why we have 
not yet stopped using the DIRMC.  With that being said more and more of my 
customers are implementing file device classes or VTL's which keep most of the 
primary data on disk.  As a result I no longer see the need for separating out 
the directories to another disk location.

About a year ago many of my larger customers would complain about how long the 
DIRMC disk pools would take to backup.  In working with support we found that 
this was a WAD feature. I think the issue was (can't remember for sure) that 
each file in a disk pool is evaluated on every backup where sequential access 
pools are evaluated differently.  As a result we started taking our DIRMC pools 
and giving them a small pool built with a file device class definition.  We 
made sure the data migrated from disk to file device class each day and as a 
result the storage pool copy problem went away.

Now the fact that we are using file device classes as described above is why I 
am concerned about the issue that was mentioned in this thread about the larger 
default block size.

All of these issues together lead me to believe that DIRMC pools are no longer 
as necessary as they used to be.

Kyle


Paul Fielding <paul AT FIELDING DOT CA> wrote:
Hi Richard,

I took a look through the Quickfacts (something I should have done long
ago). It does indeed suggest that surrogate directories are created and the
real directories are restored as they are hit.

Has anyone really observed this to be genuinely true? I have in the past
observed the double-tape-mount theory, and though I understand it is
supposedly fixed, I haven't heard anyone say "I have seen it, I know it
works, you no longer need to keep a dirmc diskpool".

Of course, if it is indeed working as designed now, it doesn't resolve the
other dirmc issues currently being discussed in this thread.

Is there anyone on the list who has in recent history decided to ditch using
a dirmc diskpool altogether and done so with success on the restore side?

regards,

Paul

----- Original Message -----
From: "Richard Sims"
To:
Sent: Saturday, March 19, 2005 4:44 AM
Subject: Re: [ADSM-L] DIRMC - Are copypool reclamation performance issues
resolved or not.


> Paul -
>
> This generally falls under the TSM term Restore Order processing. We've
> discussed it on the List before. I have an entry on it in ADSM
> QuickFacts which you can refer to as a preliminary to further pursuit
> in IBM doc.
>
> Richard Sims http://people.bu.edu/rbs
>
> On Mar 19, 2005, at 3:06 AM, Paul Fielding wrote:
>
>> I'd be interested in more discussion on this point. My original
>> understanding was actually a bit different that that. The impression
>> I had
>> was that originally directory tree structures were restored before any
>> files
>> happened, period. Following that, files would be restored. Net result
>> -
>> tapes might get mounted twice.
>>
>> Is my understanding incorrect? (could well be). If this behavior has
>> indeed
>> been fixed so that directories are restored as they are hit on the tape
>> (with a pre-created non-ACLed directory being created first) then it
>> would
>> indeed make sense that a DIRMC pool is no longer needed.
>>
>> Is there any documentation on this somewhere I can reference?
>


                
---------------------------------
Do you Yahoo!?
 Yahoo! Small Business - Try our new resources site!