Re: Testing Various Data Models?

"Mike Nicewarner" <psyclo_at_nospam_datamodel.org> wrote in message news:<c2vh8t$aab$1_at_news.netins.net>...
> I've participated in data modeling competitions before, both organizing and
> participating.
> Realize that data modeling is more of an iterative process than a concrete
> science. As the Zippy discussion thread pointed out, just because one
> modeler would win doesn't necessarily mean that the modeling tool/style was
> inherently better than the others.
> I'd be curious to hear more about what you are talking about, and I would
> love to see your reasoning (not intuition) for stating that the relational
> model should be retired.

playing usenet catchup while on the road...

Yes, I recognize that this is similar to any bake-off competition where it very well could be that the best recipe doesn't win. Perhaps:
1) This particular instance of using the recipe is less than the norm 2) The judges were looking for qualities that do not translate into getting the best recipe, but merely the best of some attribute (they might be big on texture, for example)
3) "Best" is too subjective or cannot be generalized from this one example.

Also, even if the judges do pick "the best" that doesn't necessarily lead to better industry acceptance -- if the judges choose a snake dish as the best, that will not get everyone to start eating snake, for example. But a film that wins the oscar for best picture is likely to have more people check it out and form their own opinion.

It isn't easy to figure out how to get emperical data comparing implementations of data models, much less the models themselves, but I think it is important to pursue. We have read about the mathematics of the relational model, but where is the scientific evidence of its usefulness?

It is not that the relational model gives us nothing, but a model is simply a metaphor and the relational metaphor "seems" not nearly as rich as the web metaphor, for example (the web being better modeled as a di-graph). Relational theorists typically consider "navigation" and hierarchies as bad but such concepts are apparently not bad enough to keep the web of data that forms the www from evolving as a highly useful huge distributed data repository.

Many of my concerns with the relational model could be issues with the implementations of the relational model because of the brittle nature of many applications built on current RDBMS's. That is where I have no emperical data but many snippets of anecdotal evidence. Intuition is not the same as randomly choosing an opinion. That is why I'm trying to get some facts -- because after years of experiencing a variety of information systems techniques & tools, both my gut and my pocketbook tell me that the PICK approach (which is old, I realize, but as a data model it is revived with XML) has a lot to offer the industry.

Relational proponents often give the impression that their model is beyond reproach because it is based on mathematics. I might believe in a triune God, but I sure wouldn't claim that the reason is because we can come up with a mathematical model/metaphor for such a theory. Other theories can have mathematical models too. I checked around in the Pick world and Henry Eggers told me that his guess was that if anyone had worked on putting a mathematical model to PICK it would have been him and he had done very little in that regard. So, I've done a little work on that, simply to show that "relational" is not the only approach that has been implemented for which there is a mathematical model.

Additionally, relational theorists will often agree that other models could also be based on mathematics, but they make the not-at-all mathematical statement that it is somehow clear to them that relations are the simplest construct with which to model data so it is clearly the best. [I'm a bit tired right now so the following might be a poor analogy -- but are the unitarians obviously right since they can "model God" with a geometric point where the trinitarians model God with a triangle?] When looking at a proposition or predicate, it is more likely that a person would model it with a tree (remember diagramming sentences?) than with relations. Is that relevant? I'm not sure, but I think so.

So, yes, I, too, want something more than intuition for my claim that it is time to retire (or at least significantly enhance) the relational model as it seems to have an undeserved king-of-the-hill reputation. Any insights from others are definitely appreciated. Cheers! --dawn Received on Fri Mar 19 2004 - 04:07:59 CET