Help on tuning update query

Question

Hi,
I work on below version of oracle.
SQL> select * from v$version;
BANNER
Oracle Database 10g Enterprise Edition Release 10.2.0.5.0 - 64bi
PL/SQL Release 10.2.0.5.0 - Production
CORE    10.2.0.5.0      Production
TNS for IBM/AIX RISC System/6000: Version 10.2.0.5.0 - Productio
NLSRTL Version 10.2.0.5.0 - ProductionI have below the table structure and index created on the table. I have given the update and it's execution plan also. Can you please suggest a way to tune this query
Table has close to 50 million records in it.
Thanks much in advance for helping
SQL> CREATE TABLE TABLE1
  2  (
  3    COL1    CHAR(6 BYTE)                        NOT NULL,
  4    COL2    NUMBER(13)                          NOT NULL,
  5    COL3    CHAR(2 BYTE),
  6    COL4    CHAR(3 BYTE),
  7    COL5    CHAR(5 BYTE),
  8    COL6    CHAR(7 BYTE),
  9    COL7    NUMBER(4),
10    COL8    NUMBER(6),
11    COL9    NUMBER(9),
12    COL10    NUMBER(6),
13    COL11    CHAR(1 BYTE),
14    COL12    CHAR(2 BYTE)
15  )
16  TABLESPACE AGGREGNAT_DAT_2012T4
17  PCTUSED    0
18  PCTFREE    5
19  INITRANS   1
20  MAXTRANS   255
21  STORAGE    (
22              INITIAL          80K
23              NEXT             1M
24              MINEXTENTS       1
25              MAXEXTENTS       2147483645
26              PCTINCREASE      0
27              BUFFER_POOL      DEFAULT
28             )
29  NOLOGGING
30  COMPRESS
31  NOCACHE
32  NOPARALLEL
33  MONITORING;
Table created.
SQL> CREATE INDEX IND_TABLE1 ON TABLE1
  2  (COL1, COL3, COL4, COL5, COL6,COL7)
  3  NOLOGGING
  4  TABLESPACE AGGREGNAT_IDX_2012T4
  5  PCTFREE    5
  6  INITRANS   2
  7  MAXTRANS   255
  8  STORAGE    (
  9              INITIAL          64K
10              NEXT             256K
11              MINEXTENTS       1
12              MAXEXTENTS       2147483645
13              PCTINCREASE      0
14              BUFFER_POOL      DEFAULT
15             )
16  NOPARALLEL
17  COMPRESS 6;
Index created.
SQL> explain plan for UPDATE TABLE1 o1
  2        SET o1.COL12 = 1
  3        WHERE COL4 <> 'MDD'
  4         AND EXISTS
  5        ( SELECT 'E' FROM TABLE1 o2
  6        WHERE o2.COL3 = o1.COL3 AND o2.COL4 = o1.COL4
  7                    AND o2.COL6 = o1.COL6 AND o2.COL7 = o1.COL7 AND o2.COL1 = o1.COL1
  8        AND  ( o2.COL9 < o1.COL9 OR ( o2.COL9 = o1.COL9 AND o2.rowid < o1.rowid))AND o2.COL5 = o1.COL5);
Explained.
SQL> select * from table(dbms_xplan.display);
PLAN_TABLE_OUTPUT
Plan hash value: 3057860847
| Id  | Operation                     | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | UPDATE STATEMENT              |           |  9868K|   705M|   126M  (1)|422:03:56 |
|   1 |  UPDATE                       | TABLE1    |       |       |            |          |
|*  2 |   FILTER                      |           |       |       |            |          |
|*  3 |    TABLE ACCESS FULL          | TABLE1    |    25M|  1808M| 87011   (4)| 00:17:25 |
|*  4 |    TABLE ACCESS BY INDEX ROWID| TABLE1    |     1 |    71 |     5   (0)| 00:00:01 |
|*  5 |     INDEX RANGE SCAN          | IND_TABLE1 |     1 |       |     4   (0)| 00:00:01 |
PLAN_TABLE_OUTPUT
Predicate Information (identified by operation id):
   2 - filter( EXISTS (SELECT 0 FROM "TABLE1" "O2" WHERE "O2"."COL7"=:B1 AND
              "O2"."COL6"=:B2 AND "O2"."COL5"=:B3 AND "O2"."COL4"=:B4 AND "O2"."COL3"=:B5 AND
              "O2"."COL1"=:B6 AND ("O2"."COL9"<:B7 OR "O2"."COL9"=:B8 AND "O2".ROWID<:B9)))
   3 - filter("COL4"<>'MDD')
   4 - filter("O2"."COL9"<:B1 OR "O2"."COL9"=:B2 AND "O2".ROWID<:B3)
   5 - access("O2"."COL1"=:B1 AND "O2"."COL3"=:B2 AND "O2"."COL4"=:B3 AND
PLAN_TABLE_OUTPUT
              "O2"."COL5"=:B4 AND "O2"."COL6"=:B5 AND "O2"."COL7"=:B6)
Note
   - dynamic sampling used for this statement
27 rows selected.Thank you very much for guiding
Edited by: user12869307 on Feb 6, 2013 8:52 PM
Edited by: user12869307 on Feb 6, 2013 9:09 PM
Edited by: user12869307 on Feb 6, 2013 9:11 PM

Answer 1

Well,
it depends. The query is still a bit confusing to me.
You are trying to update rows having col4 different than 'MDD' if exists another row having same value for (COL1, COL3, COL4, COL5, COL6,COL7) and having a lower or equal value of col9.
What are the business requirements? Do you have some sample data and can explain the logic of this update?
This one:
( o2.COL9 < o1.COL9 OR ( o2.COL9 = o1.COL9 AND o2.rowid < o1.rowid))can actually be written in this way
(o2.COL9 <= o1.COL9 AND o2.rowid < o1.rowid)but why using o2.rowid < o1.rowid?
.>
You should not trust on rowid order.
Check what Tom Kyte is saying in Fetching last record from a table
the rowid is not "generated" in as much as "derived"
The rowid is the address of the row on disk.
Since space can and will be reused, rowids can and will be "reused"
A rowid implies NOTHING other than the location of the row physically. It does not imply age, order of insertion, or anything like that. Regards.
Al
Edited by: Alberto Faenza on Feb 6, 2013 5:10 PM