Re: Holdable cursors in Oracle?

Billy Verreynne wrote:

> DA Morgan wrote:
>
> > The Teradata architecuture is one where while a cursor is reading a table
> > ... suppose someone comes along and wants the same data. Rather than
> > opening a second cursor with associated parsing, overhead, etc. Teradata
> > starts delivering records from the existing cursor to both sessions ...
> > and when the first user gets what they want ... the cursors just keeps
> > going with the records the second person requested. It is very efficient.
>
> Within an OLAP context, yes.
>
> I do not think the parsing is such a huge overhead myself. I'm using a
> scripting engine for a speech interface application I write. To parse and
> compile a script (containing the event handlers for the speech recognition
> engine) is usually just a few ms. And that is a way lot bigger than any
> huge and complex and nested SQL statement for a cursor. :-)
>
> What can be expensive is building and maintaining state data. In an OLAP
> environment, there's less state data to maintain in comparison with OLTP.
>
> So yeah, Teradata can likely get away with a feature like that... but
> personally I'm very hesitant in thinking that a similar feature will be of
> any real benefit in Oracle. If state data is an issue in something like
> Teradata, requiring globally shared cursors... that IMO points to an
> attempt to overcome limitations in design and not providing real froody
> features for developers and users.
>
> --
> Billy

It has been about four years since I worked with Teradata so I can't compare current Teradata vs current Oracle. But at the time ... for massive databases ... terabyte size ... you just couldn't beat Teradata for what it was originally intended to be ... a repository for terabyte sized databases. One must remember that their architecture is MPP (massive parallel processing) and the typical machine I had was 32 cpus. I suspect that advantage has all but disappeared since then but I can't say that for sure.

Daniel Morgan Received on Thu Mar 20 2003 - 10:40:41 CST