Archives of the TeradataForum

Message Posted: Wed, 14 Nov 2002 @ 00:46:47 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Duncan,

You wrote:

I don't agree this with. WHen we talk about Dimensional Models, we're talking about relational databases. A MULTIdimensional model, if you can find it, lives in a multidimensional database like Express, Essbase, Powerplay, etc. I say "if you can find it" because they aren't really models, they are mixed up with the data and you cannot abstract the model from the instance.

I'd also make the distinction for a very important reason - the DM is relational and, hence, inherits all the power of a relational model. The models inside these MOLAP tools are much more limited, almost laughable. That's why we call them cubes, it isn't meant as a compliment. They have no attributes, they only have independent dimensions. In fact, about all the information you can encode into a dimension is the name of the member and who its parent is, no attributes. Many allow only one roll-up per dimension, can't handle ragged or split hierarchies, multi-parent children or cross-dimensional attributes, roles or qualities. All of these can be handled easily in DM.

I think I've just explained why it isn't trivial, but that's most likely because most RDB practitioners don't really understand the dimensional model.

Then please characterize for me, if a data warehouse schema is NOT OLTP, but IS 3NF, but is NOT dimensional, then just what is it? What differentiates it from the two polar opposities? I've never heard this disclaimer before, that a 3NF EDW schema is not an OLTP schema. Can you be specific? (And I'm not being sarcastic, I'd really like to know)

You clearly don't know the DM, or you could not make this outrageous claim. I've been modeling RDB's for 20 years, and using the dimensional model for 12 (there aren't many people still alive who can make that claim), and I can assure, it is NOT simple and I've modeled some very complex organizations with it. But rather than make that claim, perhaps you can prove your point with a situation that is too complex for the DM.

What is a "sole" DM???? Would you make the same claim about a "sole" 3NF EDW model?

Experience would suggest otherwise. I've modeled telcos with DM. Some of those models are 7 or 8 years old, holding up very nicely.

OK Duncan, this is where you've really put your foot in your mouth. You just said that high data volumes, such as a telco, lead to dependent data marts. I presume you mean that that is a good thing. But why is it? Because, the 3NF EDW data warehouse is TOO BIG TO USE (excuse me Teradata people, I realize that this isn't as much of a constraint as it is in the other databases). So what you do is build this big database, because you think that's what you ought to do, but then you create this huge mess of downstream data marts, different schema which means more ETL and more maintenance, for users to query.

There are DM warehouses out there in the 10's of terabytes. Don't believe the claims that it can't be done.

-NR

Neil Raden
Hired Brains, Inc.
www.hiredbrains.com

	<Prev	Next>		<<First	<Prev	Next>	Last>>