ADSM-L

Re: Bufpoolsize and cache hit percentage on ADSM DB

2001-01-11 09:57:19
Subject: Re: Bufpoolsize and cache hit percentage on ADSM DB
From: Tab Trepagnier <Tab.Trepagnier AT LAITRAM DOT COM>
Date: Thu, 11 Jan 2001 08:57:05 -0600
Jonathan,

As others have stated you have to reset the statistics between measurements.

I'm running a setup almost identical.  My server has 512 MB and our DB is 22 GB,
85% utilized.

I have found that a DB buffer pool of 96 MB gives me consistent hit rates of
98.5% or better.

I'm a bit amazed that you don't have paging problems with a 220 MB DB Buf Pool.
Using default AIX memory tunings, my paging rate skyrocketed when my DB Buf Pool
exceeded 64 MB!  I shared a lot of my experiences on tuning AIX for ADSM with
this forum (check the archives).

Run vmstat with a 2-5 second sample time, and while it's running, submit a few
new queries.  If you get anything more than a brief spike on disk paging, you're
paging the contents of the DB Buf Pool to disk.  That's a performance killer,
and you'll have to either reduce its size or tune the AIX VM settings; or both.

Good luck.

Tab Trepagnier
ADSM Administrator
Laitram Corporation











"Shalev, Jonathan" <jonathan.shalev AT INTEL DOT COM> on 01/11/2001 06:05:34 AM

Please respond to "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>

To:   ADSM-L AT VM.MARIST DOT EDU
cc:    (bcc: Tab Trepagnier/Corporation/Laitram/US)
Subject:  Bufpoolsize and cache hit percentage on ADSM DB




I am running ADSM server 3.1.2.40 on AIX 4.3.2.0 with 512MB memory.
The ADSM DB size is 19GB, with 67.2% used.

I am trying to tune the Bufpoolsize so as to reach >98% cache hit
percentage, as recommended by Tivoli/IBM.

When I started, a few months ago, the value of bufpoolsize was 16K and we
had 75% cache hit %. By increasing bufpoolsize, I managed to get over 95% at
170000K.

Since then, I have had the following results (making changes every few
weeks):

Bufpoolsize = 170000 , cache hit % = 95.28.
Bufpoolsize = 190000 , cache hit % = 94.77
Bufpoolsize = 220007 , cache hit % = 93.26

All results are after many millions of DB accesses. A UNIX admin has told me
that the machine has no memory paging problems.

Does anyone have any idea why this is happening?
Am I getting better performance with the larger buffer pool even if the
cache hit % is lower?

Please cc me on list replies as I subscribe to the digest version.

Thanks,

Jonathan  Shalev
IDC Computing / Systems Engineering
UNIX Server Platforms
Intel Israel (74) Ltd.
Jonathan.Shalev AT intel DOT com
Phone: +972-4-865-6588, Fax: +972-4-865-5999
<Prev in Thread] Current Thread [Next in Thread>