Query optimization..yet again...

Question

Guys,
I need some advice yet again for tuning my query.
The query below takes hell lots of time for execution(earlier ther performance was very much appreciable).
SELECT 'B'||','||FD.GEO_ID||','||FD.PROD_ID||','||','||','||','||','||','||SUM(FACT_AMT_18-FACT_AMT_16)||','||NVL(FD.ISO_CRNCY_CODE_CHAR, 'USD')||','||FD.DUE_PERD||','||FD.MEASR_ID content
FROM AGG_FACT_605 FD
WHERE DUE_PERD = TRUNC(LAST_DAY(ADD_MONTHS(SYSDATE,-3))) + 1
AND FACT_TYPE_CODE='A8'
AND FD.PROD_ID IN (SELECT DISTINCT DSEND_ID FROM PROD_ASSOC_DNORM WHERE CTRL_PERD='20-SEP-06' AND STRCT_CODE='722' AND PARNT_ID='000000011' AND DSEND_LVL=8 )
AND FD.GEO_ID IN
(SELECT DISTINCT DSEND_ID FROM GEO_ASSOC_DNORM WHERE STRCT_CODE='705' AND CTRL_PERD='02-OCT-06' AND DSEND_LVL=6
MINUS
SELECT DISTINCT DSEND_ID FROM GEO_ASSOC_DNORM WHERE STRCT_CODE='705' AND CTRL_PERD='02-OCT-06' AND DSEND_LVL=6 AND PARNT_ID='948')
AND FD.MEASR_ID IN
('900005','900015','720000','770000','900038','900039','900031','900030')
GROUP BY FD.GEO_ID,FD.PROD_ID,FD.ISO_CRNCY_CODE_CHAR,FD.DUE_PERD,FD.MEASR_ID
ORDER BY FD.GEO_ID,FD.PROD_ID,FD.ISO_CRNCY_CODE_CHAR,FD.DUE_PERD,FD.MEASR_ID;
| Id  | Operation                   |  Name                  | Rows  | Bytes | Cost  | Pstart| Pstop |
|   0 | SELECT STATEMENT            |                        | 10546 |   628K|   377K|       |       |
|   1 |  SORT GROUP BY              |                        | 10546 |   628K|   377K|       |       |
|   2 |   HASH JOIN                 |                        |  9558K|   556M|   284K|       |       |
|   3 |    VIEW                     | VW_NSO_2               |    89 |   801 |    28 |       |       |
|   4 |     MINUS                   |                        |       |       |       |       |       |
|   5 |      SORT UNIQUE            |                        |    89 |  1869 |       |       |       |
|   6 |       PARTITION RANGE SINGLE|                        |       |       |       |   KEY |   KEY |
|   7 |        INDEX FAST FULL SCAN | GEO_ASSOC_DNORM_IDX2   |    90 |  1890 |     1 |   KEY |   KEY |
|   8 |      SORT UNIQUE            |                        |     1 |    26 |       |       |       |
|   9 |       PARTITION RANGE SINGLE|                        |       |       |       |   KEY |   KEY |
|  10 |        INDEX FAST FULL SCAN | GEO_ASSOC_DNORM_IDX3   |     1 |    26 |     1 |   KEY |   KEY |
|  11 |    HASH JOIN                |                        |  6766K|   335M|   284K|       |       |
|  12 |     VIEW                    | VW_NSO_1               |  5273 | 52730 |   235 |       |       |
|  13 |      SORT UNIQUE            |                        |  5273 |   190K|   235 |       |       |
|  14 |       PARTITION RANGE SINGLE|                        |       |       |       |   KEY |   KEY |
|  15 |        INDEX FAST FULL SCAN | PROD_ASSOC_DNORM_IDX3  |  5297 |   191K|   189 |   KEY |   KEY |
|  16 |     PARTITION RANGE ITERATOR|                        |       |       |       |     1 |     2 |
|  17 |      TABLE ACCESS FULL      | AGG_FACT_605           |    23M|   936M|   284K|     1 |     2 |
[\pre]
I beleive the problem is due to full table scan on 'AGG_FACT_605' table which contains about 1395678439 records.
If I create a partitioned table and then make use of the above query on the partioned table, what would be the best bet of column to have the partition on?
Can I've partition on combination of column values?
Any other advice to tune my query?
Help me out...
Bhagat
Message was edited by:
        Bugs                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   

Answer 1

First off you must have current stats gathered -- as has already been noted.
Beyond that there are a couple of simple things that you can do.
1) You don't need the DISTINCT clause in your sub-queries -- and it may be getting in the way of the optimizer making a bit better guesses.
2) You don't need the MINUS in your 2nd sub-query, you can use (parnt_id IS NULL OR parnt_id='948') instead.
Now if you're still seeing a full table scan on your fact table you need to ask what approach would be better than a full table scan. How selective is (i.e. what proportion of the data corresponds to) each of your predicates? In particular, how much of the data corresponds to:
1) DUE_PERD = TRUNC(LAST_DAY(ADD_MONTHS(SYSDATE,-3))) + 1
2) FACT_TYPE_CODE='A8'
3) FD.PROD_ID IN (blah)
4) FD.GEO_ID IN (stuff)
Hopefully at least one of those eliminates a significant proportion of your data such that an index can be used to retrieve only the relevant rows.