Query for performance tunning

Question

Hi,
Millions of records are there in this order table and it takes hours to run the below query
update orders
set s_last_name = decode(s_last_name, null, null, (select last_name from cust where cust_id = orders.cust_id)),
credit_card = null;
Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
UPDATE STATEMENT | | 35M| 1388M| 319K (3)| 01:03:55 |
UPDATE | ORDERS | | | | |
PARTITION RANGE ALL | | 35M| 1388M| 319K (3)| 01:03:55 |
TABLE ACCESS FULL | ORDERS | 35M| 1388M| 319K (3)| 01:03:55 |
TABLE ACCESS BY GLOBAL INDEX ROWID | CUST | 1 | 14 | 3 (0) | 00:00:01 |
INDEX RANGE SCAN | CUST_PK| 1 | | 2 (0) | 00:00:01 |
is it possible to write in a different way to improve the performance.. thanks. Bcj

Answer 1

You're running an update with an uncorrelated subquery. These can be horrifically slow. You generally have 2 options:
1. Correlated subquery
2. Do the update inside a loop. #1 if possible will probably be faster
1. Correlated subquery looks something like
update table A
    setl (column1, column2) = (select column1, column2
                                              from table2
                                            where table2.key = A.key
                                     )which might be faster, if the lookup table is properly indexed on the correlated column
2. Update in loop looks something like
  for record in (key, select column1, column2 from table)loop
    update table2
        set column1 = record.column1, column2 = record.column2
      where table2.key = record.key
  end loop;which may be faster than the original update (or may not). This method has the advantage of allowing periodic commits; people who say you should never do periodic commits have never blown out rollback segments. If you must use the loop method commit at least 4k rows (more rows per commit is better).