Re: CBO irregularity

  • From: Robyn <robyn@xxxxxxxxxxxxx>
  • To: oracle-l@xxxxxxxxxxxxx
  • Date: Tue, 8 Jun 2004 23:06:30 -0400 (EDT)

Has anyone noticed differences in CBO performance between 9.2.0.3 and
9.2.0.5?

I'm working on a data warehouse db on HP-UX (data warehouse) and I'm
finally getting to move it off of rule and on to CBO.  I've run good stats
and performance has been much better on the test server.  Still have to
get some additional proof for management, but we're on our way ...

However, we hit a memory leak bug, I'm installing patchset 9.2.0.5
right now and there have been lots of install issues.  Reading through
docs as I was creating new tars, I came across a reference to decreased
CBO performance on 9.2.0.4 and 9.2.0.5 in forum discussions.

Is there any validity to these suggestions?  I'd rather punt right now and
apply the one-off patch for the memory leak.  Our major issue on the
data warehouse has been lots of db sequential file read waits and I see a
huge decrease in the number of waits when I switched to choose mode.
This shop apparently got bit by CBO once upon a time, and if I can't
demonstrate a big performance improvement, this little effort will be DOA.

tia ... Robyn

On Tue, 8 Jun 2004, Niall Litchfield wrote:

> Well it may be restricting your database from using better execution
> plans - in fact given proper care and attention for a few key cases it
> almost certainly is.
> 
> It, if implemented properly and with the full set of required hints
> (i'd expect to see ordered in there as well for example), can help
> with plan stability if you don't want execution plans to be changed,
> some (not me) see this as an advantage. I don't. i'm not smarter than
> the CBO and I don't know what every table and its data distribution
> looks like currently let alone what it will look like in 6 months
> time.
> 
> On the other hand it is probably a sign that your developers have
> thought about good execution plans - and that would be a first for me.
> 
> Finally, do you have reason to think its bad? do your end-users curse
> the performance of the ERP system? in particular areas? then it might
> (read probably) be and is certainly worth looking at in conjunction
> with the vendor. If the performance is fine and the business is
> running OK find something else to worry about dev/QA/live for example
> :)
> 
> 
> 
> 

----------------------------------------------------------------
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: