RE: Insert contention on RAC
- From: Tim Gorman <tim@xxxxxxxxx>
- To: "Baumgartel, Paul" <paul.baumgartel@xxxxxxxxxxxxxxxxx>
- Date: Wed, 25 Jun 2008 15:06:13 -0600
Ouch! Sorry didn't see that; thanks for pointing it out...
In that case, I'm wondering if you have an index on this table
populated with a
monotonically-ascending data value (i.e. SYSDATE or a sequence
NEXTVAL). If so,
freelists certainly wouldn't help you...
Quoting "Baumgartel, Paul" <paul.baumgartel@xxxxxxxxxxxxxxxxx>:
Tim,
I'm way ahead of you! ;-) Take another look at my post; I did
pre-allocate one large extent per instance. Unfortunately, it didn't
help.
I'm pretty sure that commits are batched; they're sending a batch of
insert statements from Java and probably committing after each batch.
Transaction rates are only about 1.5 per second.
If I get down to log file sync, I'll be happy. The block contention is
what's driving me nuts. And it occurs to me that if I restrict the
inserts to one instance, I'll just get buffer busy waits instead of gc
buffer busy waits.
Help!
Thanks for your reply.
--
//www.freelists.org/webpage/oracle-l
Other related posts:
