Tim, I"m not so sure that they are always dropped cleanly (smells like a bug) win32, 10.1.0.3 Std Ed. The test case is 1 app tablespace of interest, 4MB uniform extents LMT, 6 datafiles of 1 GB each 2 tables, each an IOT according to dba_segments, slightly less than 4 GB is allocated to the 2 segments. according to dba_free_space, there is less than 20 MB free for the entire tablespace. using direct load against 2 IOTs, single load process. Dark matter, perhaps? I did see some sys-owned segments referred to in v$session_longops like "SYS.ORA_TEMP_1_nnn" but those may have been for gathering of statistics and not for maintenance of the (PK) index. Paul On Tue, 05 Oct 2004 22:18:32 -0600, Tim Gorman <tim@xxxxxxxxxxxxx> wrote: > Direct-path or APPEND inserts do avoid undo, except for the space-management > stuff in the data dictionary. > > Direct-path loads data into TEMPORARY segments which, upon successful > completion of the load (a.k.a. commit), are converted into DATA segments. > Upon unsuccessful completion of the load (a.k.a. rollback), the TEMPORARY > segments are simply dropped. > > Look ma, no undo... > > > > > on 10/5/04 9:33 PM, John Clarke at jclarke@xxxxxxxxxxxxxxx wrote: > > > /*+ append */ disables redo, not undo (as does =5Fdisable=5Flogging). > > > > Array/bulk inserts will reduce the amount of undo you generate, so this = > > method will help. But you can't avoid undo altogether. > > > > At least you're not updating or deleting the 10,000,000 rows. > > > > John > > > > ----- Original Message ----- > > From: Juan Carlos Reyes Pacheco <jreyes@xxxxxxxxxxxxxxxx> > > To: oracle-l@xxxxxxxxxxxxx > > Sent: Tue, 5 Oct 2004 20:23:01 -0400 > > Subject: RE: Please a parameter to disable undo, like =5Fdisable=5Floggi= > > ng > > > > > >> thanks > >> =20 > >> but /*+ append */ too bypasses undo, and if you set =5Fdisable=5Floggi= > > ng you run > >> the same risk, if you have a problem. > >> I think for a test database could be acceptable. For example I'm tryin= > > g to > >> create a 10 000 000 records in my home for testing > >> purposes, so this will be greatly welcome. > >> I bet there should be a way. > >> =20 > >> =20 > >> Juan Carlos Reyes Pacheco > >> OCP > > -- > //www.freelists.org/webpage/oracle-l > -- //www.freelists.org/webpage/oracle-l