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

Home -> Community -> Mailing Lists -> Oracle-L -> Re: Physics of the FILTER operation within SQL_PLANE.

Re: Physics of the FILTER operation within SQL_PLANE.

From: Jonathan Lewis <jonathan_at_jlcomp.demon.co.uk>
Date: Tue, 29 Jun 2004 09:44:34 +0100
Message-ID: <017a01c45db5$4dea7010$7102a8c0@Primary>

I've just done a little more work based on your test case, checking which rows in the filtering table get hit. Based on these observations, I think Oracle creates a hash table for the results of the FILTERing probe, using the probe "key" as the basis for the hashkey. On each probe, Oracle saves the result in the hash table, so long as there are no hash collisions. If a new probe key collides with an existing probe key, then the result is not saved. I also infer that in your test case, the size of the hash table is 128.

I would guess, but have not yet confirmed, that the size of the hash table is set as the query starts, - but only when running the cost based optimizer, so that Oracle has an estimate of the number of probe keys that need to be stored.

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

FILTER operation effectiveness depends on how (in which order) rows are inserted into driving table. Looks similar to clustering factor in index range scans ;)
Take a look on simple TESTCASE I have made on 9.2.0.4 Win2000. As you can see LIO count defers by 17 times (First case 63 LIO, second 1061) depending on order how rows have been inserted. One more effect, if we reduce row count in filter table (third test), then Oracle execute filter operation more effective (LIO=43) independing of inserting order, due to "remembering results of previous probes" probably.

  1. Any comments?
    >> [Jonathan Lewis] However, FILTER can be much more efficient than nested
    loop, because it can remember results of previous probes into the second table - effectively making the probe an in-memory lookup.
  2. Any ideas how many "results of previous probes" Oracle can "remember" for next comparison?

Best regards,
Jurijs



Please see the official ORACLE-L FAQ: http://www.orafaq.com

To unsubscribe send email to: oracle-l-request_at_freelists.org put 'unsubscribe' in the subject line.
--
Archives are at http://www.freelists.org/archives/oracle-l/
FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
Received on Tue Jun 29 2004 - 03:41:18 CDT

Original text of this message

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