Re: A question for Mr. Celko
Date: Tue, 20 Jul 2004 21:39:57 GMT
Message-Id: <pan.2004.07.20.21.40.40.244103_at_REMOVETHIS.pandora.be>
On Tue, 20 Jul 2004 02:32:15 +0000, Marshall Spight wrote:
> "Jan Hidders" <jan.hidders_at_REMOVETHIS.pandora.be> wrote in message
> news:pan.2004.07.19.22.47.17.805097_at_REMOVETHIS.pandora.be...
>> On Mon, 19 Jul 2004 11:38:57 -0700, John Jacob wrote: >> >> Just extracting them is not the problem.
>
> D'oh! Talk about an anti-climax!
>
> Could you perhaps tell us, pretty please, what the problem *is*?
I'll try. Since I'm a bit short on time, it will be brief and abstract.
The problem is that you start mixing two levels of computation. The first level is the Relation level where you manipulate the relations, probably with relational algebra or something. We have a reasonable grip on how to optimize that. The level below that are the values in the fields, the level of the Domain values. For this level separate algebras (or calculus or whatever) could and should be defined by the user. At this level serious optimization is very very difficult, especially in the case of the user-defined types. However, usually this is not a problem because you are dealing with either small things or relatively efficient operations, so non-scalability is not so much of an issue there.
Now, if you allow the UNNEST and NEST operations you are lifting values in fields to the level of the relations, and vice versa. Apart from the fact that these operations make your algebra more complicated and harder to optimize, you are now mixing the two levels and introducing the difficulties of the Domain level into the Relation level.
- Jan Hidders