We have a system which suffered from this problem, in addition to a few others. After an analysis of the what the application was doing and the requirement of consistent (not necessary 100% optimal) performance, we decided to:
*._like_with_bind_as_equality=TRUE *._optim_peek_user_binds=FALSE
All issues are gone now. I never liked bind peeking, in it's current form.
If what Jonathan says is true, we may have to revisit this decision. But so far the consistency is great, and the performance is consistently very good.
Often "always good" is prefered over "mostly awesome, but sometimes really bad".
-- Christo Kutrovsky Senior Database/System Administrator The Pythian Group - www.pythian.com I blog at http://www.pythian.com/blogs/
. > Of course this extra overhead means that this method is not suitable for ultra frequently used quick OLTP queries - but this isn't where we do have the bind variable / peeking issue anyway. . . > > Tanel.
Here I don't follow the reasoning on the exclusion.
B. Polarski
-- //www.freelists.org/webpage/oracle-l
-- Christo Kutrovsky Senior Database/System Administrator The Pythian Group - www.pythian.com I blog at http://www.pythian.com/blogs/ -- //www.freelists.org/webpage/oracle-l