Re: redo log generation and undo_retention

  • From: "Daniel W. Fink" <optimaldba@xxxxxxxxx>
  • To: oracle-l@xxxxxxxxxxxxx
  • Date: Thu, 13 May 2004 12:27:00 -0600

Josh,

Undo entries are always written into the redo stream. This is the only 
way to reconstruct undo/rollback segments in order to perform the 
rollback tasks of recovery. Undo is always 'inserted', never 'updated' 
nor 'deleted' (this is a good way to think about it, though it is not 
100% accurate). When an existing block is used by a new transaction, the 
block is basically renewed. (again, not 100% accurate, but close enough 
for 99% of the issues). The difference in undo_retention settings will 
be how long Oracle attempts to keep undo entries from being overwritten. 
As far as Oracle is concerned, undo entries into a new block or reused 
block are pretty much the same.

Daniel Fink

Jos wrote:

>Dear List,
>I am trying to find out if setting undo_retention to non-zero value will 
>generate more redo log or not. I dummy  up some update and insert transactions 
>and ran it in a 9.2 db with undo_retention=0 and undo_retention=900 (the 
>default), I monitored the redo size by querying 'redo size' stat in v$sysstat. 
>And I found out that the amount of redo generated under both undo_retention is 
>about the same. I would have thought that with undo_retention=non-zero value 
>will generate more redo because of the undo entries and these undo will be in 
>the redo stream as well. However, from the test result it doesn't seem to be 
>the case. Is it my understanding of undo is wrong or the test I did is not 
>right?
>Josh
>  
>
----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to:  oracle-l-request@xxxxxxxxxxxxx
put 'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
FAQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------

Other related posts: