Rollback taking excessively long time

To one and all:

 

Environment: ORACLE v11.2.0.3 running on AIX. Database is 120+TB, AIX has 60 CPUs behind it.

 

I ran a rather simple update against a table with 12 partitions, and about 600,000,000 rows. Had the UPDATE succeeded, it would have updated about 47,000,000 rows (modifying a date attribute from NULL to a more correct value)

For reasons of time - UPDATE had been running for over 7 hours, and looked to be about only 20% done, so I killed the UPDATE - the idea being to probably break up the UPDATE into more manageable chunks and retry bit by bit.

 

Problem: DB is rolling back - slowlyyyyyy. When I monitor V$TRANSACTION, I saw that there were originally 280,000 undo blocks, and I have calculated that it is rolling back at the rate of about 5,000 blocks/hr - good old math says 56 hours to complete.

 

Question: Why would it take so bloody long to rollback a transaction that had run for only 7 hours? There was only one table being modified?

 

Is there some  DB setting that, if set outrageously wrong, would cause this slowness?

 

BTW, I am not the system DBA, only the app DBA, but the system DBA's are clueless.

 

Also, I can live with the 56 hours, because this table is used only weekly, and we have 96 hours till the next job run.

 

Any ideas would be appreciated.

Stephen Miller - email: miller_stephen at usa.net