01 December 2007

DBWR and the Devil

I was recently re-reading parts I, II, III and IV of Kevin Closson's series on the perils of overconfiguring your db_writer_processes (which I mentioned earlier this year). My instance appeared to be a perfect candidate. It was configured with db_writer_processes=4, using the one-DBWR-for-each-CPU approach that Kevin rails against for most configurations.

After some other reading and investigation, I decided to make the change to only use one DBWR processes. Changes to the db_writer_processes parameter require an instance bounce to take effect, and we had some downtime scheduled anyway.

I'm happy to say that there are no ill effects. I don't the see DBWn process anywhere on the top(1) radar. I can't say I haven't seen any drastic performance improvements, but if I can do the same work at the same level with 1/4 of the resources, I'd call that a win. It's probably a bigger win for our development/test server that had 3 production clones, each with 4 DBWR processes, for a total of 12 DBWRs. This is now 1 each for 3 total, and (again) no reported performance hit.

No comments:

Post a Comment