Java Mailing List Archive

http://www.dba.5341.com/

Home » Home (12/2007) » oracle-l »

RE: os cache vs. db cache

Milen Kulev

2007-07-10

Replies:

Nachricht
Hi Robin,
I have got positive experience using  os cache when running OLTP (not DWH!) databases.
I believe it was a "special" case -> many concurrent sessions were accessing the same index (hot) blocks.
We have got many "cache buffer chains" latches, and they constituted ~ 20-25% of database resource usage (top5 wait events as of STATSPACK reports). After  swithing on the os cache (HPUX 11.11, 16 CPUs, Veritas Filesystem ), the "cache buffer chains" issues vanisched.
So the overhead of double buffering was in this case offset by elimination of latch contention.
Response time for all the critical queries halved.  
HTH. Milen 
 
P.S.   It was third party application, where creating storedoutlines after each application release was not fun at all ;(
 
-----Original Message-----
From: oracle-l-bounce@freelists.org [mailto:oracle-l-bounce@freelists.org] On Behalf Of Robyn
Sent: Monday, July 09, 2007 8:04 PM
To: oracle-l
Subject: os cache vs. db cache

Hello everyone,

Over the years, I've read differing opinions on balancing the os cache and the database cache.  Seems like I remember there were some advocates for minimizing the os cache and maximizing the database cache to allow Oracle to better determine which blocks should truly be cached and avoid necessary blocks being cached because they were in proximity to hot data.  I also have heard opposite opinions.

I'm specifically focused on performance for a data warehouse.  Has anyone tested performance with a minimized OS cache and a maximized database cache?  If so, what were the results?  Does anyone have a really good paper or book on the topic?  I'm trying to collect some information before I start discussions with the platforms guys.

tia ... Robyn
©2008 dba.5341.com - Jax Systems, LLC, U.S.A.