Re: Nested structures

From: dawn <dawnwolthuis_at_gmail.com>
Date: 16 Sep 2006 08:24:36 -0700
Message-ID: <1158420276.655283.112860_at_k70g2000cwa.googlegroups.com>


JOG wrote:
> dawn wrote:
> > Some, but not all, of you will find this new IDC paper entitled
> > "Because Not All Data is Flat: IBM's U2 Extended Relational DBMSs" to
> > be of interest.
> > [snip]
>
> How about this then - I propose that all data is, in fact, very much
> flat.

The word "flat" leads to unnecesary confusion, but if I'm interpreting your use of it correctly, then I would say that all data can be modeled with 1st order logic. OK, I just read your next sentence, so yes, I agree.

> All information we communicate to each other may be stated as a
> proposition in first order logic. Everything.

Yes, and wouldn't that be a delightful way to communicate ;-)

I agree, and frankly, even if there were some exception, I am quite comfortable saying that anything I've ever worked on could have been modeled with first order logic.

We could also say that most data can be modeled with structures that could be visualized in 3-dimensional space or with monadic second order logic too, therefore not flat.

So, all data could be modeled as flat and all data could also be modeled as full-figured (or whatever term you want to use for not-flat, "multidimensional" is another term that causes confusion). Note that trivial or simple, aka flat, data would look flat within the multidimsional structures. I think the author is calling this data that would look flat even if you had the option of multivalued attributes, for example, as flat data and other data as "not flat."

When we glance around at unmodeled propositions, some look flat without further manipulation (by this use of the term "flat") and some do not, right? All can be modeled by flat structures, however.

> Hence, any collection of information may be organized into sets of
> propositions, as determined by commonalities in the structure of those
> propositions. The information world could be no flatter.

Agreed. It is the next leap that many make where the problem comes in. That not-exactly tight argument goes something like this "all data can be modeled with 1st order predicate logic, 1st order logic is simpler than 2nd order logic, therefore we should model all data by putting it in the format formerly known as first normal form"

There are too many flaws in that one, including issues that I now know relational theorists admit were flawed in the earlier RM efforts, but that most practitioners are still stuck working with. (e.g. 1NF by the understanding held by most practitioners who studied database theory in the past 20 years, and still the norm for any products supporting SQL-92 or ODBC)

Restating it with the word "flat", the relational mantra seems to be something like this:

All data can be modeled as flat, therefore all data should be modeled as flat.

That argument is not exactly mathematical, but we've been down this route before. smiles. --dawn Received on Sat Sep 16 2006 - 17:24:36 CEST

Original text of this message