Oracle-L: RE: Indentifying Redundant Indexes

From: Cary Millsap <cary.millsap_at_hotsos.com>
Date: Sun, 22 Dec 2002 21:58:39 -0800
Message-ID: <F001.00520FEF.20021222215839@fatcity.com>

Mark,

The way we do it is by what you might call extreme brute force:

Capture all SQL from the applications (not just the stuff you've run in the past month, but the stuff you'll run in the future too, like period-end close processes, and so on).
Generate execution plans in production for all this SQL. Store the plans.
In a "structural replica" system (that is, a full-schema test instance; you don't need real application data, but you do need schema and db statistics imported from the production system), drop the index.
Generate execution plans on the replica system for all the SQL. Store the plans.
Compare the two sets of plans from steps 2 and 4.
Decide whether the different in 4 that are different from the plans in 2 are better or worse than the plans in 2.

Like I said, it's a "big hammer" method, but it has its reliability advantages, and the only step that we haven't automated is #6. (I'm assuming that you already have a valid test system as described in #3.) The tool we use that does steps 2, 4, and 5 is called Project Laredo (www.hotsos.com/products/laredo).

Cary Millsap
Hotsos Enterprises, Ltd.
http://www.hotsos.com

Upcoming events:

- Hotsos Clinic 101, Jan 7-9 Knoxville
- Steve Adams's Miracle Master Class, Jan 13-15 Copenhagen
- 2003 Hotsos Symposium, Feb 9-12 Dallas

-----Original Message-----
Richard
Sent: Sunday, December 22, 2002 10:54 PM To: Multiple recipients of list ORACLE-L

Dear List,

Firstly - Merry Christmas to those who have not already departed for the holiday season.

I'm currently doing some investigation based around indexes and would like
everyone's opinion: What is everyone's preferred approach to identify redundant (as in "never used by Oracle") indexes? I believe Oracle 9 might
have a feature to set a flag on objects and then check back later to see if
they have been accessed however we're still stuck on 8.1.7.4 (Solaris). Some of my thoughts include:

Can query for physical disk i/o at a tablespace level easily, however more difficult to go to an object level.
Could create a trace file and then inspect explain plans for existence of index accesses, however trace file probably not practical to capture for a long period of time.
Ideal statistic would be something along the lines of "index x used y times in last 24 hours", however a simple "index x was used in the last 24 hours" would be ok.

Obviously we are searching for indexes to remove and identifying those which aren't queried over a set period of time would be good candidates for
a starting point. Any advice you might have would be greatly appreciated.

Regards,

Mark.

<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<---->>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

>>>>>>>>

Privileged/Confidential information may be contained in this message.

          If you are not the addressee indicated in this message
       (or responsible for delivery of the message to such person),
            you may not copy or deliver this message to anyone.

In such case, you should destroy this message and kindly notify the sender

by reply e-mail or by telephone on (61 3) 9612-6999. Please advise immediately if you or your employer does not consent to

                Internet e-mail for messages of this kind.
        Opinions, conclusions and other information in this message
              that do not relate to the official business of
                         Transurban City Link Ltd
         shall be understood as neither given nor endorsed by it.
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<---->>>>>>>>>>>>>>>>>>>>>>>>>>>>>

>>>>>>>>

-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.net
-- 
Author: Mark Richard
  INET: mrichard_at_transurban.com.au

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).

-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.net
-- 
Author: Cary Millsap
  INET: cary.millsap_at_hotsos.com

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).

Received on Sun Dec 22 2002 - 23:58:39 CST