Comments on: Heterogeneous Data Layers? Check. What’s Next?

By: stephen o'grady

stephen o'grady — Tue, 02 May 2006 02:51:42 +0000

G. Roper: thanks for the comment. apparently i wasn’t clear, but fortunately Andy’s explained my point better than i could. when i talk about heterogeneous data layers, i’m not trying to define some new term - i’m merely talking about the trend of application developers to employ multiple persistence approaches side by side, as it were. relational, as you point out, is one of them more often than not for the reasons that you mention.

but it’s equally true that for some problems - partitioning, e.g., relational models are often less than ideal.

are they going away? of course not. are they still hugely relevant? you bet.

but i think we’re beginning to see a relaxation in the “relational = the one and only acceptable approach.” when you talk about databases, incidentally, i would also be sure to include filesystems as another mechanism.

Andy: thanks very much for clarifying; spot on.

By: Andy Fundinger

Andy Fundinger — Mon, 01 May 2006 19:14:07 +0000

I read it as heterogenous as in employing more than one storage system in the data layer. A simple, common, case being a file system for bulk storage and a database for metadata.

By: G. Roper

G. Roper — Sun, 30 Apr 2006 05:30:22 +0000

When using a new term like “heterogeneous data layers” it is useful to define the term. All I can infer is that the term means “non-relational”.

Abandonment of the relational model usually means either abandonment of expectations (”we don’t have enough money to do this right”, “we don’t need perfection - 80% correct will do.”) or lack of understanding of the relational model (”Normalization? We don’t need no stinkin’ normalization!”). In some applications (e.g., CAD) an OODB may be more appropriate. Even in those cases there is inevitably a follow-on requirement for an interface to a relational database, and so much of the work defining a relational database must still be done.

I want to ask: for these non-relational “data layers” where will the semantics (meaning) of the data be defined? How will relationships between data elements be defined/enforced? Will you use pointers or unique keywords to point to the children of a parent?

Until someone produces an artificial intelligence product that can think, has multiple domain knowledge and can negotiate meaning with other similar entities, then there will be no database “silver bullet”.

Meanwhile we know what’s available today: relational, hierarchical, network, and object-oriented databases. I don’t see anything revolutionary around the corner. But vendors will enhance their products to address the changing markets.