Re: Optimizer change from 8i-9i

Autotrace uses explain and that can produce a different plan from the run time plan whenever bind variables are involved. That is documented in the performance manual.

Christian referred to the indes tie-break rule as well in his first reply. If there are two suitable indexes, the RBO chooses the one with the higher object_id (i.e. the newer one) - or is it the other way around? In any case, if as part of a migration you drop and recreate indexes you could and up with different plans even with the RBO if the indexes are not created in the same order they were before.

zhu chao wrote:
> Hi, Chris,
> Thanks for your answer.
> I remember autotrace does not print the plan correctly, when
> histogram is used under certain conditions. But I am not sure now and
> I will do some test to verify it.
> And autotrace also have limit print the partition prune. (In your
> case and partition prune, it did not print the wrong plan , just with
> some limitation actualy, right?)
>
> Maybe other casae that it reported the wrong plan, I will keep it
> updated when I am aware of it.
>
>

-- 
Regards

Wolfgang Breitling
Centrex Consulting Corporation
www.centrexcc.com
--
http://www.freelists.org/webpage/oracle-l