Normalized Data base

I have been arguing with my partner if we shoud or not design a normalized
data base or denormalized. How, can we know when we should choose one or the
other?
Thanks a lot!
Lina,
I assume you are referring to designing a data warehouse since you are
posting in this group. There are three basic ways of designing the model.
Third Normal Form - This is the Inmon approach and works well data mining
and building down stream data marts. It's draw backs are usually higher
development, hardware, and software costs. It's should be used for large
data warehouses with multiple sources.
Star Schema - A special from of denormalized schema. This is the Kimbal
approach. It good in situations of combining multiple sources and the use
is intended to be more direct user access.
Denormalized - This should be used in data warehouses that really only have
one significant source and the warehouse will be used from more direct
reporting than anything else. It's relatively quick simple and cheap but
has issues with integration.
Whichever approach you choose to take realize that once reports and other
uses are written against it, changing the general approach is expensive.
Just opinions...
"Lina Manjarres" <LinaManjarres@.discussions.microsoft.com> wrote in message
news:A3CF8EA4-FAB4-4831-8BBF-4AFBCD41F5C6@.microsoft.com...
>I have been arguing with my partner if we shoud or not design a normalized
> data base or denormalized. How, can we know when we should choose one or
> the
> other?
> Thanks a lot!
|||I guess it depends on what type of denormalization. The more denormalized a
datawarehouse model is, the harder it will be to keep the data consistent as
new data is inserted or updated incrementally from the transactional
database. For example, if a person's zip code changes or a product is
associated with a new category, these changes must be updated in multiple
locations. However, if the datawarehouse is reloaded from scratch
periodically (similar to a reporting database), then it is less of an issue.
"Lina Manjarres" <LinaManjarres@.discussions.microsoft.com> wrote in message
news:A3CF8EA4-FAB4-4831-8BBF-4AFBCD41F5C6@.microsoft.com...
> I have been arguing with my partner if we shoud or not design a normalized
> data base or denormalized. How, can we know when we should choose one or
the
> other?
> Thanks a lot!
|||Hi Lina,
if you go to my downloads page on www.peternolan.com there are some
presentations, one of which is about data modeling and some newsletters
one of which is about data modeling.
There is something of a 'reglious' debate over modeling techniques in
DWing that has been going on for many years. I put my public opinion on
my FAQs page...
Current 'best practice' of publicly available information is to have an
archive layer in 3NF + Time variant model (plus stability analysis if
you need it) as well as an analytical later (dimensional). This is also
the most expensive practice......not everyone needs an archive layer
as described here...
The 'next generation' models I work with make the separate archive
layer redundant...I've written on this topic many times and the posts
are archived on http://www.datawarehousing.com/list.asp.
Best Regards
Peter Nolan
www.peternolan.com
sql