I believe its Mogens chapter in the Tales of the Oak Table book where he said he found with 10g that LIOs and CPU usage do not necessarily correspend. He argues that tuning queries should be explicitly based on elapsed time.
My understanding of LIOs is that every LIO is a buffer cache latch get, so even if you do not use up more CPU you are incurring serialization and under concurrency can cause performance problems. I have seen queries go from 20,000 LIOs down to 300 with a very small performance improvement. Is it worth it to spend the time to do this?
BTW, its a very good book. The chapter by Dave Ensor on the history of Oracle is one of the best chapters you can find anywhere. I hope he writes more now that he is retired.
--
To unsubscribe - mailto:oracle-l-request_at_freelists.org&subject=unsubscribe
To search the archives - http://www.freelists.org/archives/oracle-l/
Received on Tue Sep 07 2004 - 08:20:16 CDT
