Re: Frequent log switches during sqlldr batch jobs

OEM has some performance testing to document if there are performance problems. I can't decide if they are really useful, although if there are redo latch waits that is perhaps meaningful.

When I have systems that have a great variance in redo usage at different times, I usually size the redo for the high usage times, and cron a switch to keep low usage times at reasonable intervals (which probably means I'm old-fashioned more than anything else regarding init parameters), where reasonable is defined as management-approved recovery granularity.

I think as long as you've followed reasonable rules of thumb about how to configure redo and archiving, and you aren't getting any errors in the alert log, and aren't having issues with getting things done in load windows, you don't have a problem. Another consideration might be number of archived logs for RMAN (and in the controfile if you are nocataloging), and the pain of manually going through huge RMAN logs simply because of a high number of filenames.

I'm sure you've considered nologging options followed by a backup? (see direct-path insert in concepts manual). If you get huge archives from the load, this can be worth it. Especially if it means you can replace extraction, moving and loading with CTAS.

jg

--
@home.com is bogus.
"ah, delayed block cleanout for unix :-)" - Connor McDonald