Re: are LIOs always relavent?

Not an answer to your question directly but, perhaps, you can try to add one filtering condition to your join query so that it targets specific partition (you mentioned that all rows in GTT match a single partition anyway).

Also, do you do any significant amount of parsing? I noticed that parsing on partitioned objects is somewhat heavier than non-partitioned (well, at least in 9i). Since you don't have much difference in CPU consumption - it's probably not the issue.

Perhaps, useful to rerun it with dynamic sampling 1 or 0 and without bind variable peaking and check if the difference is still there.

I would also check execution plans and see if join order is the same or look for whatever other variations besides "partition list iterator". Maybe you can try to collect statistics only globally on table and not on partition level.

2006/7/1, ryan_gaffuri_at_comcast.net <ryan_gaffuri_at_comcast.net>:
> The real question is... How can I have a large spike(sometimes double) in LIOs, have the same physical IOs, no additional waits, and at the same time have similiar CPU usage and response time?

-- 
Best regards,
Alex Gorbachev

http://blog.oracloid.com
--
http://www.freelists.org/webpage/oracle-l