Lists Home |
Date Index |
10/28/02 11:23:01 PM, "Rick Jelliffe" <firstname.lastname@example.org> wrote:
>rather are a physical model. Consequently, most schema languages for
>XML will work best (if one is trying to use them as a logical model) for
>data sets when there is little difference between the logical and the physical.
>?Which kinds of data are these: trivially, heirarchical structures with
>a strong ordering between elements. And, structures which can typically
>be processed efficiently by knowledge of the ancestors and their attributes.
This ties in with what somebody wrote about RDBMSs lacking sufficient performance for chip-design
applications and the like. In chip design and general electronic design, the majority of your
information is contained in the spatial and temporal ordering of your primitive elements rather
than the values of the elements themselves. The same is true for a lot of "semi-structured
data." The problem is that the relational model, by explicit design, doesn't concern itself with
ordering; relations are *sets*, dammit. This means that the ordering information has to be
encoded and decoded by explicit application logic, which takes time and space (for the data
elements whose only purpose is to "thread" others).
To a large extent, the relational model's ability to abstract away ordering concerns has been its
big strength when dealing with the sort of data in which ordering issues were historically matters
of internal implementation detail rather than essential features of the data itself. But that
strength turns into a weakness when dealing with the type of data where order carries (most of
the) genuine information. This is why neither the relational model nor XML nor anything else can
fulfill the role of a Grand Unified Model (GUM); the entire purpose of a model is to abstract away
the "irrelevant" features of whatever it is that's being modeled, and what is or is not relevant
depends on the problem domain.