| Pete Dashwood 2007-10-16, 6:55 pm |
|
"Howard Brazee" <howard@brazee.net> wrote in message
news:onm9h35iiq416288cvb4ec32cmed1mse6d@
4ax.com...
> On Thu, 11 Oct 2007 22:13:57 -0500, Robert <no@e.mail> wrote:
>
>
> "Normal form" can be a religion. Do users care about Codd's rules as
> long as they can have reliable data?
>
> Actually users like data warehouses. How many data warehouses have
> everything normalized?
Although I am quite devout about the Relational Model, I agree with you
Howard.
Experience in the real world (and with some very smart people who have made
RDB their life's work) has shown that, although normalization is generally a
"Good Thing" there are cases where unnormalized datasets can provide better
performance. As you observed, this is particular true for Data Warehousing,
where pivot tables and cubes which can switch quickly between
multidimensional data views are more important that normalized relations.
Often these tables are built dynamically "on-the-fly" so normalizing them
would be difficult anyway. The underlying base structures can benefit from
normalization, just like most RDBs, but the "coal face" (especially n OLAP
environments) doesn't necessarily need to be.
Pete.
--
"I used to write COBOL...now I can do anything."
|