Subject: RE: Please a parameter to disable undo, like _disable_logging

"Goulet, Dick" wrote:

>There are some things like this that I believe should not be done, even
in test. The >reason is that now you've crippled a significant item for database recovery for the >purpose of shaving a couple of microseconds off of the execution time. Problem with that >is that when the same application starts to run in production it will run slower than in >test & damanagement will want the same performance/crippled database in production. Then >when, not if, all hell breaks loose you're the one trying to explain why the database >can't be recovered. I personally like having my test environment, not development, in a >similar configuration so that all of the background delays play out just like in
>production.

Recovery is the key word. The recovery requirements for a test system are different than for OLTP production or other types of production (except when testing recovery, of course). I have a situation where the gigantic data files are reloaded every night nologging - we're talking shaving hours, not microseconds. As far as those are concerned, the recovery granularity is a day. But that db also contains Portal and SSO stuff, which is transactional. So I do nightly hot backups with the gigantic files readonly - again, minutes versus hours. Everything else is DSS so I don't care about recovering it, they just have to do it over if there were ever a crash. Of course, test/development isn't separate from production here, impossible to justify _another_ 3 servers just for me. The frustrating thing is Portal generating over 100M per day archived logs, even when no one is doing anything.

But you are right in that damagement perception trumps all. And that a test system should accurately reflect a production system. Personally I would like giving developers their own systems, and having test, QA and production systems successively more tightly controlled. But I never seem to get that.

http://www.garry.to

--
http://www.freelists.org/webpage/oracle-l