Archives of the TeradataForum

Message Posted: Fri, 08 Nov 2002 @ 18:01:02 GMT

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

Anomy Anom,

While I tend to agree with Mr. McBride and Ms. Myers, I wouldn't base this decision solely on performance. (this coming from a software company the sells a 3NF SQL generator)

Where is your data coming from, how long does it take to get the data into shape to load, how long to you expect the load to take, and how often are you loading? This is very important!

Generally, a Dimensional model takes more "up-front" effort to get the data in shape for loading (increased time and complexity).

Dimensional models make querying easier (less complex SQL), at the expense of query flexibility (Mr. McBride and Ms. Myers make this case very effectively).

3NF models typically are easier to load, but the SQL required is more complex. The use of complex views, to hide data model complexity, tends to hamper query performance, which is one of the reasons why 3NF models get a bad rap.

Depending on what your raw data looks like, I would lean towards 3NF. Load the data, run some queries and see if your tools can effectively handle the query complexity. It is easier, over time, to move toward a "Dimensional" model, if you find it absolutely necessary.

John Grace
Geppetto's Workshop

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