Materialized View Initial Creation vs Complete Refresh - MAJOR differences

Question

Hi All,
I am currently in the process of setting up MV's for replication via a DBLink and have noticed something that I find quite odd. Basically If I create an MV I get v. large differences than when I do a complete refresh on the same MV after the initial "load". Actual stats are shown below :
Action TimeTaken Undo Reqd (Bytes)
Initial Load 00:01:07     37254998
Complete Refresh     00:02:41     463919012
Question is why does a complete refresh require 12x the amount of Undo than the creation of the MV? I though tit would be roughly the same or at most twice as much?
The Instance is only being used by myself so no other processes are causing this difference and this can be reproduced. Undo calculated from V$ROLLSTAT.
Thanks
MH

Answer 1

Hi tekicora,
Thanks for your help and suggestions. I have tried this method of evaluating UNDO with the following results:
STATISTIC# NAME SID VALUE
176 undo change vector size 115 132
CREATE MATERIALIZED VIEW mytest1
TABLESPACE TABSPACEA NOLOGGING BUILD IMMEDIATE
USING INDEX TABLESPACE TABSPACEB REFRESH FORCE WITH ROWID ON DEMAND AS
select * from ptable@remdb;
STATISTIC# NAME SID VALUE
176 undo change vector size 115 332860
exec DBMS_MVIEW.REFRESH('MYTEST1', 'C',ATOMIC_REFRESH=>TRUE);
STATISTIC# NAME SID VALUE
176 undo change vector size 115 261460092
exec DBMS_MVIEW.REFRESH('MYTEST1', 'C',ATOMIC_REFRESH=>FALSE);
STATISTIC# NAME SID VALUE
176 undo change vector size 115 272282756
This is still creating masses of undo (particulalry atomic refresh). Could this be because mv table source is on a remote db?
Thanks
Mark H
Message was edited by:
markyb0y