Re: An oldie on Blocksize and Performance?

In that strange way that life has of imitating the internet, I've just come off a site which had a performance problem -

They wanted to join a 25M row table to a reference table, then crunch the result down to a 1.5M row summary.

The fastest (obvious) solution took 900 CPU seconds, but hit the cache buffers chains latch 100M times, with 9.9M misses. The next solution should have gone faster, but the latch miss rate went up from 10% to 12.5%.

Currently though, we have plan C which is a little more subtle, but completes in 10 minutes on its own, but 12 minutes average if we do 4 of them concurrently.

No wonder I got so interested.

BTW - Have you looked at the parent latch for CACHE BUFFERS CHAINS - I don't think it should be taken before a child, but maybe it is, and maybe that's where you are suffering most.

    (v$latch_parent).

--

Jonathan Lewis
Yet another Oracle-related web site: www.jlcomp.demon.co.uk Received on Thu May 13 1999 - 03:44:57 CDT