Re: question about automatic undo management

"Howard J. Rogers" <howardjr2000_at_yahoo.com.au> wrote in message news:<s6LX9.30783$jM5.79038_at_newsfeeds.bigpond.com>...
> "Joel Garry" <joel-garry_at_home.com> wrote in message
> > Consider a humongous DW combined with a humongous transactional labor
> > and inventory system with humongous RBS's and product planners running
> > humongous part and labor allocation what-if scenarios for building
> > jumbo-jets on it. For each what-if, when they are done, they
> > flashback. At some point, they are doing so much undo I/O, it becomes
> > substantially like the old contention problems.
> >
>
>
> Sorry Joel: I don't buy this. Flashback means 'show me my table data as it
> used to be'. The essence of what-iffing (is that a word? Well, it is now!)
> is that you don't monkey around with the real data, but subset it into some
> sort of temporary holding area, and do your monkeying around there. When the
> model comes out not to your satisfaction, dispose of it... there's no
> rolling back there.

Your mistake is in assuming the average person who hasn't had you for a teacher will do it right, rather than trying the hyped features and then being unhappy.

>
> I've done modelling with global temporary tables (8i and 9i), and of course
> the prize whatiffing strategy must go to Workspace Management (9i only).
> Either way, undo is generated, it's true, but I don't see either strategy
> *necessarily* straining undo I/O. And if large amount of undo are generated,
> and that starts becoming an I/O problem, then configure more undo tablespace
> datafiles and stripe them across more hard disks.

I've seen it done with dedicated hardware. I talked to one of the developers who would be handling it for a big project, and he seemed to have a reaction of "wow, I've never had a dataset that big!" No scalability problems there, eh?

jg

--
@home is bogus.
It's a wicked world we live in.