My question is about measuring, and improving, Disk performance under the
configuration described below.
We have a Alpha 4100 3GB memory 4 CPU (5/500), server, Tru64 v.4.0e,
connected to a RA7000 HSZ70 Dual Controller (Dual Redundant Mode) Raid Array
with 32 MB mirrored write enabled cache containing 6 9.1 GB 7200 RPM drives
Configured for RAID 3/5. This Array is partitioned to 5 logical drives
contain data only (The OS and Swap are on a separate SCSI bus). The server
is running the Progress Database version 8.3B, with multiple databases and
an application that accesses that data. The application is a 3rd party
product, so we don't have any control on how the underlying database tables
are structured.
During ordinary transaction processing by interactive users, the performance
is excellent. However, last weekend during an application update that
required conversion to the database structure, processing these databases to
excessively long (4 days!). I'm not a Progress DBA but as I understand it,
the database structure change is handled by Progress programs that compare
the existing DB structure to an empty database containing the new structure,
creating a "meta" file of changes. This "meta" file is then used to drive
the Progress process that resturctures the original database. It was this
process that ran I/O bound the entire time it was executing (CPU time was
near nothing, TOP showed the progresses consistently in a WAIT state). I
understand that Progress uses what it calls a "before image (BI)" file into
which it writes uncommited transactions before flushing them to the acutal
database. In this server's configuration these files are located on the
same filesystem as the database file itself.
This Raid Array about to receive an additional 7 18 GB drives to handle the
growth of the existing databases and the addition of more databases.
Management is up-in-arms about the length of time this conversion took to
complete and I am charged with finding ways to improve the performance of
the databases. This weekend I will be taking advantage of the new disks to
explore how changing the database file layout might improve this
performance, eg: comparing mirror set performance improvement over RAID 3/5,
locating the BI file onto separate disks, etc.
What I'm looking for are suggestion on 2 things.
First, how (what tools?) can I use to actually measure the throughput of the
disk array while this batch job runs? Something I can use to show
performance percentage improvement between different database/Raid Array
layout.
Second, could any experienced Progress administrators provide suggestions on
how you've laid out you databases files for best performance? The nagging
question in my head in all this is whether the Progress database made
intelligent decision in the method it took to modify the database structure
from the "meta" file. For example, did it drop secondary indexes and
recreate them at the end of the conversion process if it would have made
sense to do so?
Any suggestion will be most appreciated. If I've ommitted any information,
please let me know and I'll provide it.
Will summarize.
Thanks,
Kevin Erickson
kreleif_at_hotmail.com
_____________________________________________________________________________________
Get more from the Web. FREE MSN Explorer download :
http://explorer.msn.com
Received on Tue Dec 12 2000 - 14:40:09 NZDT