Re: Undo Generation

I believe you are using 10g.

There's an optimisation for undo and redo in 10g which aims to minimise the number of records (vectors for redo) on array processing.

For inserts, all inserts made at the same time to a single block can be packed into a undo record (ditto redo vector).

For delete, every row deleted has to generate a separate undo record (redo vector).

For updates there is an odd mix (which may also be true for inserts, except that I haven't looked closely enough). The number of undo records can be between 1 and the number of rows updated. I think this depends on the number of rows which can be updated 'in situ' - so your 9 suggests that several rows had to be transferred into the block free space for the update to take place. If you had done

    set sal = sal + 1
you may have found fewer undo records.

(I guess I ought to test whether an insert that causes deleted rows to be cleared out and the free space to be coalesced results in many more undo records, or just one or two extra - or maybe one per deleted row cleared).

Regards

Jonathan Lewis
http://jonathanlewis.wordpress.com

Author: Cost Based Oracle: Fundamentals
http://www.jlcomp.demon.co.uk/cbo_book/ind_book.html

The Co-operative Oracle Users' FAQ
http://www.jlcomp.demon.co.uk/faq/ind_faq.html

• Original Message -----
  > Date: Thu, 28 Jun 2007 23:13:39 +0530
  > From: "DBA Deepak" <oracle.tutorials_at_gmail.com>
  > Subject: Undo Generation
  >
  > Hi Gurus,
  > Have a simple doubt. I did an experiment to know which DML statement
  > generates the Most/Least amount of UNDO for the same no. of records.
  >
  > Steps:
  >
  > ===========================================================
  >
  > SQL> conn scott/tiger
  > Connected.
  >
  > SQL> create table myemp as select * from emp;
  >
  > Table created.
  >
  > SQL> delete myemp;
  >
  > 14 rows deleted.
  >
  > SQL> select used_ublk,used_urec from v$session s,v$transaction t where
  > s.taddr=t.addr and s.username='SCOTT';
  >
  > USED_UBLK USED_UREC
  > ---------- ----------
  > 2 14
  >
  > SQL> rollback;
  >
  > Rollback complete.
  >
  > SQL> update myemp set sal=sal*1.1;
  >
  > 14 rows updated.
  >
  > USED_UBLK USED_UREC
  > ---------- ----------
  > 1 9
  >
  > SQL> rollback;
  >
  > Rollback complete.
  >
  > SQL> insert into myemp select * from emp;
  >
  > 14 rows created.
  >
  > USED_UBLK USED_UREC
  > ---------- ----------
  > 1 1
  >
  > SQL> rollback;
  >
  > Rollback complete.
  > =================================================================
  > Please note that there was only one session connected to the database as
  > SCOTT user.
  >
  >
  > My question is why the do the UNDO generated for DELTE and UPDATE DML
  > statements differ? Because Oracle has to keep undo information of the same
  > no. of records in both the cases.
  >
  > Please help me on this.
  >
  >
  -- http://www.freelists.org/webpage/oracle-l