ADSM-L

Poor database performance

2003-04-01 09:51:28
Subject: Poor database performance
From: Tab Trepagnier <Tab.Trepagnier AT LAITRAM DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Tue, 1 Apr 2003 08:46:56 -0600
I just migrated our TSM system to a new pSeries 6H1 from our old F50.
While I/O throughput is much higher, the database performance is
surprisingly poor.  I am asking for help in trying to figure out why and
what I can do about it.

TSM 4.1.5.0 running on AIX 4.3.3 ML 10
6H1 2-way 600 MHz 64-bit
4 GB RAM
27 GB TSM database with 256 MB DB buffer pool
Four 4-drive libraries, each double-connected via SCSI (two drives per
adapter)

This system has been in service for one day.  The same backup system had
been running on the old F50 since ADSM 2.

On our old F50, I had split the database into many 1 GB volumes as an
experiment when we had only 512 MB in our TSM server.  It didn't seem to
hurt performance so I'd left it that way.

When I built the new server, I took advantage of info on this forum and
created a single large DB volume on each DB disk.  The DB is currently
laid out this way:

SCSI Adapter #1 --> DB disk 1 (one vol) --> DB disk 3 (one vol) --> Log
disk 1 (five vols) --> Pool disk 1

SCSI Adapter #2 --> DB disk 2 (one vol) --> DB disk 4 (one vol) --> Log
disk 2 (five vols) --> Pool disk 2

DB and log volumes on odd-number disks are TSM mirrored to volumes on the
even-number disks (the volume on DB disk 1 is TSM mirrored to the volume
on DB disk 2 for example).

Yesterday we saw decent performance during expiration and reclamation of
two primary tape pools.  When a third pool's reclamation started, and
especially when the system had to handle reclamation of all three primary
pools and the copy pool and client sessions, the overall performance was
very poor.  Watching system activity with topas showed overall disk I/O on
the DB disks to be very low - much lower than we saw on the old F50.

Oh... I know TSM 4.1 is not supported any more.  This server upgrade was a
prerequisite to the software upgrade which will occur in about a month or
so.

Also, I know I can set a larger DB buffer pool than 256 MB and will do so
over the next couple/few days.  But we had better database performance on
the F50 with a DB buffer pool of only 128 MB.

==> So what should I do?  Split the database into "a few" volumes?  Or
should I look elsewhere?

Thanks in advance.

Tab Trepagnier
TSM Administrator
Laitram LLC

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