Re: Directory storage pools

2004-04-04 14:45:43
Subject: Re: Directory storage pools
From: TSM_User <tsm_user AT YAHOO DOT COM>
Date: Sun, 4 Apr 2004 11:45:07 -0700
I thought that a long time ago a restore had to recover the directories first 
before it could recover the data. At that time implementing a separate storage 
pool for directories that stayed on disk made sense in order to speed up 

For some time now a restore will lay down a dummy directory if it hits a file 
that has to be restored before the directory is restored. As a result the need 
to have directories in a separate pool seems to be less necessary.

Still, some people still implementing this separate storage pool for 
directories when they have many windows clients. Large windows file servers may 
have millions of directory objects that will be stored in a storage pool. Just 
like backing up millions of small files if these objects are spread across many 
tapes the restore will be slower. I have seen that you can cut 5 - 10% off of 
the restore time when the directories are all on disk.

I would still use the DIRMC if you are a shop with many mgmt classes and a lot 
of windows nodes. Lets say your MGMT class with the longest retention is called 
"LONG". Your data goes to an MGMT class called "SHORT". If the storage pools 
defined in both mgmt classes are collocated then the directory data for the 
node will take up an extra tape in the storage pool defined to "LONG". Multiply 
that, times enough nodes and it can make a difference. Further, if you have 
more than one storage pool that has the longest retention then TSM randomly 
determines which one to bind the data too. Lastly, if a new mgmt class is 
created that has a longer retention then a lot of rebinding will happen.

I looked at what the redbook said I believe the redbook is speaking to an issue 
that has long since been resolved.

Richard Sims <rbs AT BU DOT EDU> wrote:
>As per the TSM redbook at:
>It suggests use of DISKDIRS and OFFDIRS to store directory information
>to avoid restore-time directory reconstruction hits.
>Everything makes sense, including the need to explicitly bind all
>directories on a client to the DISKDIRS management class with DIRMC.
>The next question is... *how* do you select ONLY directories for this
>management class without selecting any files within them, in dsm.opt?
>(Server is on AIX and clients are on Windows, Linux,
>Solaris, and AIX.)

Directories cannot be independently selected for a Backup or Archive
operation: they are implicitly backed up or archived along with files
unless you cause files only to be operated upon. Perhaps it would be
helpful if you posted the essence of what you are trying to achieve.

Note that you don't *need* to use DIRMc with Backup: you can use it for
"steering" purposes, but otherwise directories are bound to the management
class with the longest retention period. Note also that you have a mix
of Windows and Unix clients, wherein only the Windows directory info will
typically end up in a storage pool: as the redbook says, "On UNIX clients,
the directory entries are stored directly in the IBM Tivoli Storage Manager
server database" (unless ACLs are involved). With Unix directory info,
the DIRMc effects retention choice, but stgpool choice is not a factor.

I have notes on this stuff in ,
as an adjunct to the primary IBM reference material.

Richard Sims

Do you Yahoo!?
Yahoo! Small Business $15K Web Design Giveaway - Enter today

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