Maybe I missed something but, perhaps, you are not scoping your data properly in time and/or sessions. If you don't have many XCTENDs than you shouldn't have many log file sync waits either.
You said "once the big chunk of processing starts...". In this case I assume that this chunk is your business critical part you are keen to optimize. Thus, you should limit analysis only to this part.
On Tue, 10 Oct 2006, Christian Antognini wrote:
> For every commit there is a line in the tracefile. You can use a command > like the following one to count them: > > grep "XCTEND rlbk=0" <tracefile> | wc -l
hi - did that already - once the big chunk of processing starts there are no XCTENDs until near the very end. That is what is puzzling - I can't see performance problems writing to the disks and I don't see lots of XCTENDs... I do not know if XCTEND is issued when a cursor commits and the other cursors are still open doing work...
Thanks! -- http://www.freelists.org/webpage/oracle-l
-- Best regards, Alex Gorbachev
The Pythian Group Sr. Oracle DBA