Performance Issue - Index is not used when a zero padded string is queried

Question

Hi All,
I have a table T1 which has many columns. One such column say C1 is a varchar2(20). T1 has 10 million rows and there is an index called I1 on C1. Stats are current for both tables and indexes. These are the scenarios:
Scenario 1
select *   from T1 where C1 = '0013206263' --Uses index I1
187 ms
Scenario 2
select *   from T1 where C1 = '8177341863' --Uses index I1
203 ms
*Scenario 3*
*select *   from T1 where C1 = '0000000945' --Uses Fulll Table Scan --Very Slow*
*45 seconds*
When I force the sql to use the index through a hint, it is working fine:
Scenario 4
select /*+ INDEX (t1 i1) */  *   from T1 where C1 = '0013206263' --Uses index I1
123 ms
Scenario 5
select /*+ INDEX (t1 i1) */  *   from T1 where C1 = '8177341863' --Uses index I1
201 ms
*Scenario 6*
*select /*+ INDEX (t1 i1) */  *   from T1 where C1 = '0000000945' --Uses index I1*
*172ms*Is there any reason for this performance issue? Why does the optimizer goes for FTS in Scenario 3?
Edited by: user539954 on May 14, 2009 12:22 PM
Edited by: user539954 on May 14, 2009 12:32 PM

Answer 1

user539954 wrote:
Please see the replies below:
- How many distinct values for C1 out of that 10 million rows? I'm guessing that histograms were created for C1, correct?
=>7 million distinct c1 values. I have not gathered a histogram yet. Should I try that?
SQL> explain plan for select * from T1  where C1 = '0000000954';
| Id  | Operation         | Name           | Rows  | Bytes | Cost (%CPU)|
|   0 | SELECT STATEMENT  |                |   244K|    19M| 26228   (5)|
|   1 |  TABLE ACCESS FULL| T1 |   244K|    19M| 26228   (5)|
SQL> explain plan for select * from T1  where C1 = '0033454555';
| Id  | Operation                   | Name               | Rows  | Bytes | Cost (%CPU)|
|   0 | SELECT STATEMENT            |                    |   532 | 43624 |   261   (0)|
|   1 |  TABLE ACCESS BY INDEX ROWID| T1     |   532 | 43624 |   261   (0)|
|   2 |   INDEX RANGE SCAN          | I1 |   532 |       |     2   (0)|
It's possible you do have a histogram, even though you didn't plan on creating it, if you're running 10g.
In the absence of a histogram and with 7M distinct keys in 10M rows, Oracle should have predicted 2 rows for both queries, not 244,000 and 532.
If you do have a histogram, you probably need to get rid of it.
Regards
Jonathan Lewis
http://jonathanlewis.wordpress.com
http://www.jlcomp.demon.co.uk
"Science is more than a body of knowledge; it is a way of thinking"
Carl Sagan
To post code, statspack/AWR report, execution plans or trace files, start and end the section with the tag {noformat}{noformat} (lowercase, curly brackets, no spaces) so that the text appears in fixed format.