ADSM-L

Re: Database backup performance

2001-06-27 13:50:53
Subject: Re: Database backup performance
From: Bill Colwell <bcolwell AT DRAPER DOT COM>
Date: Wed, 27 Jun 2001 13:34:33 -0400
Zoltan,

I have a very large os/390 server.  The last full dbb went at a rate of
3.3 million pages/hr, almost 5 times fater than yours.

Since v1 r1 of adsm I haven't found any tricks to speed this up
except to change hardware.  My current system is

cpu: 9672-ra6
tape: 9840 running 3590 emulation
disk: STK SVA.

5 years ago the dbb went at a rate of 510,000 pages an hour.
The hardware then was STK iceberg disk and 3490 tapes, I don't
remember the processor.

I don't think dbb speed has anything to do with the cache hit%.
A sequential process like a full dbb should stay out of the cache since
it would otherwise flood the cache with pages that aren't likely to be
referenced again.

Hope this helps,

--
--------------------------
--------------------------
Bill Colwell
Bill Colwell
C. S. Draper Lab
Cambridge, Ma.
bcolwell AT draper DOT com
--------------------------
In <OFDE139284.A5C35858-ON85256A78.0057D3F2 AT vcu DOT edu>, on 06/27/01
In <OFDE139284.A5C35858-ON85256A78.0057D3F2 AT vcu DOT edu>, on 06/27/01
   at 01:34 PM, Zoltan Forray/AC/VCU <zforray AT VCU DOT EDU> said:

>I guess I should have clarified that the server is on OS/390, not *NIX !!

>Thanks for all the responses on how long it takes. Now, how about ways to
>improve the speed ?

>I am trying a backup to 3590 drives, to see how much of a difference it
>makes. So far, there does seem to be a speed difference.

>I also went through the ADSM job log and collected some statistics.

>The last FULL DB backup took more than 5-hours for 3665050 pages !!!

>*******************************************************************************
>>>> Hello, we full backup our 22.3 GB DB (assigned, used 92%)
>in 39 minutes, using Magstar 3590E cartridges in the 3494 robotic.
>Th DB is laying on mirrowed ssa-disks (7133) using 18GB-disks
>(ibm7133-8518)
>Processor in RS600/H50, OS/AIX 4.3.3 , TSM4.1.3.0

>...mabe you should first check the capability of the disks where your db
>is located - what you would get there by reading on unix for example
>a simple test would be done using dd for example
>( if you have your server on unix)
>Example:
># time dd if=/some_data_located_on_the_db_disk of=/dev/null bs=1024k
>count=100
>100+0 records in.
>100+0 records out.

>real    0m8.396s
>user    0m0.010s
>sys     0m1.890s
>#
>... would get the time to just read 100MB ( just at the time of the test )
>running the example -dd- test again can to unrealistic low times when
>data is cached, so a new test should be done with new data ( if=...)
<Prev in Thread] Current Thread [Next in Thread>