Excessive undo segment extension

Question

I am using the automatic UNDO feature in 10g.
The UNDO tablespace size is 3gb with autoextend set ON.
Undo retention is set at 900.
I am attempting to update 1M+ records contained in a global temporary table using FORALL .. UPDATE method.
About 2 minutes into the process, the performance degrades dramatically as undo segment extension alerts eventually make the application freeze.
Other io alerts ( direct path write temp, busy buffer waits, etc ) also appear.
I notice that the UNDO tablespace has an init extent of about 131k and additional extents set at 64k.
These seem small for my needs.
Can someone please advise ?
Thanks !

Answer 1

Can you check from AWR to investigate what waited event happended?
http://www.oracle-base.com/articles/10g/AutomaticWorkloadRepository10g.php
Use EM to check waited event.
If you have another session else accessed this table often, Perhaps you need commit ... often (example: commit every 100000 rows).
If you found redo log often switch log ... when you update ...
you can use "nologging" option to help
update table_name nologging set ....Anyway if you think the problem was about undo... you should found error in alert log or your session...(ora-01555)