Hi Matthias,
I don't think that SLOWINCREMENTAL will make any difference on the restore, as
this option is for backup. When SLOWINCREMENTAL YES is in effect, ADSM uses a
different algorithm for processing backups that is less memory-intensive. For
restores, though, it doesn't have any effect.
I don't know why you are seeing the same tapes mounted many times. ADSM sorts
the list of files that will be restored such that each tape is mounted once and
read from front to back. However, if you are running multiple, concurrent
DSMC.NLM processes, then each process might call for the same tape as another
process (and thus might have to wait until another process finishes using that
tape). Also, to avoid having to restore the directories first, then the data,
ADSM builds a "skeleton" directory structure first. This way, files in a
directory can be restored without having to first restore the directory (the
directory will be restored as it is encountered on the tape). Otherwise
multiple passes on the tapes would be required to restore the directories first
, then the files. The skeleton directory structure prevents this scenario.
Some things that might help restore performance (maybe you are already doing
these):
1) Unless you are in a network-constrained environment, do not use client
compression, as decompression can definitely slow things down.
2) If you have a routed network environment, do you have this line in
SYS:ETC\TCPIP.CFG:
TcpMSSinternetlimit OFF
(note that this is case sensitive, and should be entered as I've shown.)
3) Use TXNBYTELIMIT 25600 in the DSM.OPT file, and TXNGROUPMAX 256 in the ADSM
server options data set.
4) Set up a separate disk pool that does not migrate to tape, and use DIRMC to
send directory backups to it. Remember to set up the directory management class
such that it has the longest retention attributes of your other managment
classes.
5) If you want to be able to do parallel restores, it's best to plan ahead by
breaking them down by directory, and backing up each directory (and
subdirectories) to different storage pools. For example, suppose I have a
volume with three directories in the root:
DATA:MYDIR1
DATA:MYDIR2
DATA:MYDIR3
I could set up three management classes that each point to a different storage
pool, then use INCLUDE statements to back them up to the desired management
class:
INCLUDE DATA:MYDIR1\...\* MG1
INCLUDE DATA:MYDIR2\...\* MG2
INCLUDE DATA:MYDIR3\...\* MG3
Then if I have to do a restore, I can do:
LOAD DSMC RESTORE -SUBDIR=YES DATA:MYDIR1\*
LOAD DSMC RESTORE -SUBDIR=YES DATA;MYDIR2\*
LOAD DSMC RESTORE -SUBDIR=YES DATA:MYDIR3\*
Since the data in each directory is in a separate storage pool, I know that one
restore won't wait on a tape being used by another restore. And the directories
will be restored from the disk pool I set up for them.
Andy Raibeck
ADSM Level 2 Support
---------------------- Forwarded by Andrew Raibeck/San Jose/IBM on 07/11/97
10:02 AM ---------------------------
10:02 AM ---------------------------
ADSM-L AT VM.MARIST DOT EDU
07/11/97 03:26 AM
Please respond to ADSM-L AT VM.MARIST DOT EDU @ internet
To: ADSM-L AT VM.MARIST DOT EDU @ internet
cc:
Subject: Netware restore performance
Hi to all ADSM'ers
I'm just running an catastrophic restore test of on of my Netware
4.11servers.
I'm very disappointed about the restore performance.
The ADSM-Server is MVS with an database of 4.8GB and 3490e tapes,
ADSM-lvl is 2.1.10.
There are 350.000 objects with 8GB size to restore.
Most of data is on tapes with collocation on.
When starting the restore for the first time the server has been running
out of
memory, so I switched 'slowincremental' to on.
The next action was to split the restore into two pieces and running
two adsm-nlm's on the Netware-server which restore different data.
Now it's already running 20 hours and I have restored 4 GB / 200.000 files.
I'm seeing something which I can't understand :
Whenever one of the two processes has to restore an directory the other
process seems to wait, also it reads from an different tape.
There are hundreds of tape mounts and I see the same tapes
mounted many times.
I've read in the 'adsm performance tuning guide' that I should give
an mc to the directories (DIRMC) to store them on harddisk an not
on tape.
Would this help to increase the restore-performance ?
Until now I'm storing my directories on tapes with collocation on.
How can I move directories which are stored on tape back to the disk-pool ?
I understood that running with slowincremental=no would restore the
directories first and the files after this.
Would this give better performance also I've not stored my dirs on disk ?
My problem is that running with this option on crashes my server.
How do the other ADSM'ers work when the have to restore an complete server
?
Do you use other product beside ADSM ?
I'm thinking of products which are able to run an image-backup of my
server-disks
an run the incremtal restore of adsm against it.
Does anyone have experience with image-backup-tools on Netware or
how do you handle the huge numbers of files ?
I know that this is a lot of questions shortly before the weekend...
But perhaps there is anyone out there with some hints for me...
Thanks in advance,
Matthias Brasch, HEW-Hamburg Germany
m_brasch AT compuserve DOT com
|
