Re: A question for Mr. Celko

Jan Hidders <jan.hidders_at_REMOVETHIS.pandora.be> wrote in message news:<pan.2004.07.18.13.17.46.586491_at_REMOVETHIS.pandora.be>...
> That approach rests upon two assumptions: (1) you have enough useful
> algebraic identies for optimizing and (2) the algebraic expression
> corresponds roughly to a real query evaluation plan. For the PACK and
> UNPACK operators I haven't seen much proof for (1) and (2) seems doubtful
> since they don't look operators you would directly implement that way.

I have to admit I haven't spent much time thinking about the actual implementation, but I have to disagree that the primitives aren't there. The operations are well-defined in terms of existing relational algebra operators. Surely the question of optimization can be handled at the implementation level and does not affect the logical model. This sounds to me just like the arguments against join when it was first proposed. (That will never work, you'd have to compute cartesion products, etc.)

> It's an extra assumption that, from a logical point of view, is not
> necessary. I don't see how that simplifies things.

I think it is necessary, many of the operators used to define the higher level functions involved rely on the definition of intervals as discrete. For example, the MEETS operator requires that the successor and predecessor of any given value be available. Granted this can also get into some granularity discussions, but that's another story.

> And do you have a
> formal proof that you are not sacrificing power?

No, it's an admittedly intuitive notion for me, but we have been modeling time discretely in DBMS's, and in computers in general, for a long time now. I'm not even sure what a non-discrete model would look like, nor why it would allow me to represent something that I couldn't represent with discrete time. It seems like more of an academic question than a practical one.

Regards,
Bryn Received on Mon Jul 19 2004 - 21:02:23 CEST