Join cardinality and query tuning.

From: Charu Joshi <joshic_at_mahindrabt.com>
Date: Mon, 1 Nov 2004 20:51:50 +0530
Message-ID: <MHEAIPLKCACENJKNJIALMELGCIAA.joshic@mahindrabt.com>

Hi all,

Please see the plan below. The estimated cardinality of the join between the two tables is 3274K. This doesn't seem to tally with the figure I have derived from the statistics in the database. Where am I going wrong?

| 0 | SELECT STATEMENT | | 1 | 543 |
| 1 | SORT AGGREGATE | | 1 | |
| 2 | HASH JOIN | | 3274K| 543 |
| 3 | TABLE ACCESS FULL | W_LOV_D | 17 | 13 |
| 4 | INDEX FAST FULL SCAN| W_ACTIVITY_F_N5 | 3452K| 525 |

Let me try to elaborate on the problem:

The SQL statement is:

select count(*)
from
W_ACTIVITY_F T29238,
W_LOV_D T230600
where
T29238.X_BT_OUTCOME_AREA_WID = T230600.ROW_WID and T230600.VAL in ('Save-No', 'Save-Yes')

I have used the following formula for calculating the join selectivity:

JS = (1/MAX(NDV(T1.c1), NDV(T2.c1)))
     * (Card(T1) - Num_Nulls(T1.c1)/Card(T1))
     * (Card(T2) - Num_Nulls(T2.c1)/Card(T2))

Join Cardinality = SelectedRows(T1)* SelectedRows(T2)* JS where SelectedRows(T) = Card(T) * FilterFactor

(From Mr. Breitling's famous paper on 10053 event).

The database statistics are:

siebel_at_VOLANL>SELECT table_name, num_rows FROM user_tables 2 where table_name IN ('W_ACTIVITY_F', 'W_LOV_D');

TABLE_NAME                       NUM_ROWS
------------------------------ ----------
W_ACTIVITY_F                     16858500
W_LOV_D                             24000

siebel_at_VOLANL>SELECT table_name, column_name, num_distinct, num_nulls   2 FROM user_tab_col_statistics
  3 WHERE (table_name='W_ACTIVITY_F' AND column_name = 'X_BT_OUTCOME_AREA_WID')
  4 OR ( table_name='W_LOV_D' AND column_name = 'ROW_WID');

TABLE_NAME COLUMN_NAME NUM_DISTINCT NUM_NULLS

--------------- --------------------- ------------ ----------
W_ACTIVITY_F    X_BT_OUTCOME_AREA_WID           12   13406400
W_LOV_D         ROW_WID                      22720          0

So the Join Selectivity base on this information would be:

JS =(1/MAX(12,22720))*((16858500-13406400)/16858500)* ((24000 - 0)/24000)

9.0127E-06

And the Join Cardinality will be

JC = 9.0127E-06 * 17 * 3274K

528.916608

where 17 = SelectedRows(W_LOV_D) - as per the explain plan.

3274K = Card(W_ACTIVITY_F) -
Num_Nulls(W_ACTIVITY_F.X_BT_OUTCOME_AREA_WID)

Again as per the explain plan

So Oracle should have estimated it to be 529. But it has estimated it 3274K. Why?? What other factor is influencing the join cardinality estimate??

Now here's part 2:

I think the reason Oracle decides to do an FTS and Index FFS on W_LOV_D and W_ACTIVITY_F tables respectively, is because of the incorrect join cardinality estimate. The query actually returns only 67 rows!! The best execution plan for this query is obviously INDEXED RANGE SCAN of W_LOV_D followed by NESTED LOOPS join into W_ACTIVITY_F with INDEXED RANGE SCAN as the table access method. I cannot provide any hint as the front-end dynamically generates the queries. How do I manipulate the statistics so that Oracle will choose this execution plan?

Many thanks & regards,
Charu.

Disclaimer:

This message (including any attachments) contains confidential information intended for a specific individual and purpose, and is protected by law. If you are not the intended recipient, you should delete this message and are hereby notified that any disclosure, copying, or distribution of this message, or the taking of any action based on it, is strictly prohibited.

Visit us at http://www.mahindrabt.com

--
http://www.freelists.org/webpage/oracle-l

Received on Mon Nov 01 2004 - 09:18:13 CST