Re: Effect of multiple block sizes on performance
- From: "Jonathan Lewis" <jonathan@xxxxxxxxxxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Sat, 12 Jun 2004 11:55:03 +0100
There are a few little oddities where
performance can be affected.
1) Using single table hash clusters - smaller
block sizes can give a slight edge if you can't
guarantee non-collision of hash keys.
2) If you aren't using CPU costing, the cost
of tablescans is affected non-linearly, so (e.g.)
a 16 block read on 4K blocks does not have
the same cost as an 8 block read on 8K blocks.
Bigger blocksizes will introduce a bias towards
tablescans.
Of course, different block sizes may require
different numbers of block gets for the same
query, so possible changes in contention on
latching. Bigger blocks could cause more
collisions on inserts, updates, and deletes,
especially on indexes.
Regards
Jonathan Lewis
http://www.jlcomp.demon.co.uk
http://www.jlcomp.demon.co.uk/faq/ind_faq.html
The Co-operative Oracle Users' FAQ
http://www.jlcomp.demon.co.uk/seminar.html
Optimising Oracle Seminar - schedule updated May 1st
----- Original Message -----
From: <ryan.gaffuri@xxxxxxx>
To: <oracle-l@xxxxxxxxxxxxx>
Sent: Thursday, June 10, 2004 7:54 PM
Subject: Effect of multiple block sizes on performance
: We have to publish data from several staging databases and we are using
transportable tablespaces. What are the effects of multiple blocksizes on
performance?
:
: Has anyone used this under very heavy user load? We have a 4-Node RAC and
worst case stress level could be 2,000 transactions/second. We will also
have a reporting element(which I hope will only run over night).
:
: Anyone use this? Not sure I want to, just looking at it. Its a hybrid
application so scaling is rather tricky.
----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx
put 'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
FAQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
Other related posts:
