Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Mailing Lists -> Oracle-L -> rm RULE based optimizer != GOOD IDEA

rm RULE based optimizer != GOOD IDEA

From: Post, Ethan <Ethan.Post_at_ps.net>
Date: Thu, 21 Apr 2005 16:46:38 -0500
Message-ID: <E5D2567DDF0D03428A15664A78CA473C3B5D9C@pscdalpexch02.perotsystems.net>


Just got finished working on a little SQL perf problem. The query was something like...

select a.col, b.col, (select count(distinct foo) from tbl3 where batch_id=3Db.batch_id) errors
  from tbl1 a,

       tbl2 b

 where a.batch_id=3Db.batch_id
   and a.status not in ('X','Z')=20
   and b.trans_type=3D'Q';

Now that ran really slow using the COST optimizer...

but when we added the RULE hint, it ran about much faster (50 sec to < 1 sec).

Of course we can't do that so we spent a lot of time writing the query in different ways, moved the select count(*) into an inline view and joined to that, query dropped from 50 seconds to about 20 seconds but did not approach the sub-second performance of the RULE method.

The explain plan was not helpful because it looked the same as the plan generated using RULE but the difference was in the way the select count() was being handled, and the plan wasn't showing us that.

So what is the gist of this post? Well, I am sure if we continued to look at this that we might come up with a solution but why waste all that time when /*+RULE */ does the trick? What would be the point of Oracle removing something that time and time again demonstrates it's usefulness like this? It essentially (in this particular case) is a GO_FASTER hint and it worked. So maybe they will get rid of the ability to set the optimizer mode to RULE but this will hopefully always be there at the statement level.

--
http://www.freelists.org/webpage/oracle-l
Received on Thu Apr 21 2005 - 17:51:51 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US