750K Updates Daily to a 70 million table

Question

We have a pretty wide table with 50 + columns (varChar and other) that needs updating daily.
Around 250K inserts and 750K updates. Inserts are not a problem.
The ETL process doing the updates currently is very slow and I would like some ideas as to how to go about improving the performance of this load.
Adding more info:
Since, this is via an ETL process--it is doing row by row updates. I do not have an explain plan.
Database is Oracle 11g 64bit
The dimension table has 25 indexes including an index on the column I am basing the update on.
There is no partitioning on the table at this time.
I was looking for suggestions --maybe to speed up the update by using a temporary table or something. I have read about CTAS but am not sure whether that is applicable here considering there are only 750K updates.
Edited by: zebango on Mar 20, 2012 9:20 AM

Answer 1

zebango wrote:
We have a pretty wide table with 50 + columns (varChar and other) that needs updating daily.
Around 250K inserts and 750K updates. Inserts are not a problem.
The ETL process doing the updates currently is very slow and I would like some ideas as to how to go about improving the performance of this load.
Adding more info:
Since, this is via an ETL process--it is doing row by row updates. I do not have an explain plan.
Database is Oracle 11g 64bit
The dimension table has 25 indexes including an index on the column I am basing the update on.
There is no partitioning on the table at this time.
I was looking for suggestions --maybe to speed up the update by using a temporary table or something. I have read about CTAS but am not sure whether that is applicable here considering there are only 750K updates.
since the amount of data update is high i would go by following method to increase the performance:
Suppose you want to update EMP table with LOWER(ename), then
1) create table new_emp as select empno, LOWER(ename) ename, JOB, MGR, HIREDATE, SAL, COMM, DEPTNO from emp;
2) drop table emp;
3) rename new_emp to emp;
Now in step 1) i could also have used /*+ full parallel(tablename, 10) */ to increase performance further, but this depends upon the cpus.
Second option would be
Drop indexes, constraints before update and rebuild in parallel once update is done...
Third option would be
Use parallel DML
Fourth option would be
Adjust PCTFREE and PCTUSED parameters to increase performance of update
Have a look at
http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:6407993912330