ADSM-L

Performance problems

1997-09-16 15:12:16
Subject: Performance problems
From: Bryan Baars <bbaars AT CASECORP DOT COM>
Date: Tue, 16 Sep 1997 14:12:16 -0500
We recently upgraded our RS/6000 box to AIX 4.2.  Prior to this upgrade,
ADSM backups were taking anywhere from 1.5 to 2 hours to process.  Now,
however, backups are taking 4.5 hours and even longer.  The amount of
data being backed up is the same.  Here are some details about the
server/upgrade:

During the upgrade planning, IBM told us to install the following for
ADSM and our Magstar tape drive:

        Atape driver (for Magstar)      3.1.0.6
        ADSM client                     2.1.0.6
        ADSM Server                     2.1.5.13

The server is a J40 with 4 CPUs.  There is one internal drive for
rootvg.  This is where ADSM is installed, as well as the database,
recovery log, and storage pool definitions.  This is the exact same
configuration that we had prior to the upgrade (I realize storing the
db, log, and stg. pools on the same disk will cause I/O contention - but
we didn't experience these problems before...)

In the ADSM backup logs, I see the data transfer rate vary anywhere from
500KB/sec to 3,500KB/sec (yesterday's backup was around 3,500KB/sec, but
the overall backup time took over 4 hours for 432MB total transfer with
compression).  I would suspect some of the backups would take longer
because I had a trace turned on, but even without the trace, the overall
time was unbearable.  I have also noticed the system load average
fluctuate between 1.00 and 15.00.  The CPU states also show very heavy
kernel CPU usage, but then there can be 50% idle time.  iostat shows
lots of small transfer rates while ADSM backs up small files (52KB -
1000KB every 5 seconds) but more reasonable transfer rates (2MB-10MB
every 5 seconds) while backing up larger files (20MB or bigger).  The
ADSM trace files show a high time value in the Data Verb section, but I
don't know what this means and whether or not this is normal.

I have been working with ADSM support for two weeks now, and they
haven't found any reasons why this is happening.  I've tweeked all the
ADSM parms per their (IBM) recommendations.  In the last conversation I
had with them, they wanted me to install the latest ADSM server patch
2.1.5.15 and an OS bos patch that supposedly will help the process of
ADSM writing to journaled filesystems.  I am leery of just throwing
patches at my system - especially if IBM doesn't know if the patch will
fix the problem.

Has anyone experienced similar problems or does anyone have additional
suggestions to help me troubleshoot this?  BTW - please respond directly
to me (bbaars AT casecorp DOT com), I am still trying to get re-subscribed to
this list.  Thanks in advance!
--
Bryan C. Baars                          CASE Corporation
Bryan C. Baars                          CASE Corporation
UNIX Administrator                      Racine, WI
<Prev in Thread] Current Thread [Next in Thread>