How can I rewrite this update stmt to improve its poor performance?

Question

Hi,
I have the following update stmt that runs for over 6 hours. Here is the SQL and its plan:
            UPDATE TABLE1
                 SET mainveh = 'Y'
             WHERE (comp#,polnum,a8dtef,a8deef,a8dtac,
                    DECODE(everif,'Y',10000,0)+auunit*100+vrcdsq)
                IN (SELECT comp#,polnum,a8dtef,a8deef,a8dtac,
                           MAX(DECODE(everif,'Y',10000,0)+auunit*100+vrcdsq)
                      FROM TABLE1
                     GROUP BY comp#,polnum,a8dtef,a8deef,a8dtac);
PLAN_TABLE_OUTPUT
| Id  | Operation             | Name     | Rows  | Bytes |TempSpc| Cost (%CPU)|
|   0 | UPDATE STATEMENT      |          |     1 |   108 |       |   798K  (1)|
|   1 |  UPDATE               | TABLE1   |       |       |       |            |
|   2 |   HASH JOIN           |          |     1 |   108 |  1079M|   798K  (1)|
|   3 |    TABLE ACCESS FULL  | TABLE1   |    21M|   834M|       |   224K  (1)|
|   4 |    VIEW               | VW_NSO_1 |    21M|  1364M|       |   440K  (1)|
|   5 |     SORT GROUP BY     |          |    21M|   794M|  2453M|   440K  (1)|
|   6 |      TABLE ACCESS FULL| TABLE1   |    21M|   794M|       |   224K  (1)|I'm using Oracle 10.2.0.3. The TABLE1 table has 21 million rows. The update stmt will update about 15 million rows. How can I rewrite this update stmt so it'll perform better? There is a primary index on all the columns selected in the subquery. That is the only index on TABLE1.
Thank you!
Edited by: user6053424 on Jul 21, 2010 6:59 AM

Answer 1

Hi,
Thank you for your suggestions. There is an index on the columns in the group by, it is the PK index on TABLE1. I'm suspecting that due to the amount of data to update, the optimizer decided that full table scan is cheaper than index scan. I'm very interested in the GTT idea, but still need some help if I decide to create a GTT from the subquery, because if I just do this:
create global temporary table table1_tmp
on commit preserve rows
as SELECT comp#,polnum,a8dtef,a8deef,a8dtac,
           MAX(DECODE(everif,'Y',10000,0)+auunit*100+vrcdsq)
      FROM TABLE1
     GROUP BY comp#,polnum,a8dtef,a8deef,a8dtac;then the original update stmt still has the DECODE and such in it, I'm not sure how much benefit that'll do to us:
UPDATE TABLE1
                 SET mainveh = 'Y'
             WHERE (comp#,polnum,a8dtef,a8deef,a8dtac,
                    DECODE(everif,'Y',10000,0)+auunit*100+vrcdsq)
                IN (SELECT comp#,polnum,a8dtef,a8deef,a8dtac,???
                      FROM TABLE1);Your input is greatly appreciated! Thanks!