Rule based algorithm to convert an ER model to a dimensional model

View previous topic View next topic Go down

Rule based algorithm to convert an ER model to a dimensional model

Post  ubethke on Mon Oct 18, 2010 4:41 pm

I recently came across two white papers by Moody/Kortnik that show how to use a rule based algorithm to transform an ER model to a dimensional model.

I have made the papers available on my blog.

http://www.business-intelligence-quotient.com/?p=1083

This is really an interesting read. While it may not fit all of the requirements it covers a lot of ground.

In my opinion this approach could also be useful for generating virtual dimensional models, e.g. on top of your 3NF EDW or directly on top of a source application (if you really wish to do so).

I'd like to hear your opinions on the paper and if anyone has come across them before?

ubethke

Posts : 28
Join date : 2009-02-03

View user profile http://www.business-intelligence-quotient.com

Back to top Go down

Re: Rule based algorithm to convert an ER model to a dimensional model

Post  ngalemmo on Mon Oct 18, 2010 5:46 pm

I've read it before. It doesn't add much practical information that hasn't be covered elsewhere and makes a few misleading statements.

Can one mechanically create a star schema from a relational model? Sure. Will it be a good model... not likely.

The authors fail to explain how attributes on the order header, for example, fail to wind up in the data model. The dimensions of the final sales fact model are those entities that had a FK relationship with either the order header or line. Very simplistic and not complete.

There is also a presumption that the relational model itself accurately reflects the subject areas for analysis. Do we know, for example, if returns are handled by the order system and wither or not they need to be included in sales analysis? If they are, in fact, handled in a different set of tables in the operational system, how does one automajically include such information in the sales model if such information is desired?

Basically, dimensional modeling demands that you look at data from a different point of view than a relational model. Going from one to the other requires a deeper understanding of the nature of the business and its processes than what can be derived from a data model alone.
avatar
ngalemmo

Posts : 3000
Join date : 2009-05-15
Location : Los Angeles

View user profile http://aginity.com

Back to top Go down

View previous topic View next topic Back to top

- Similar topics

 
Permissions in this forum:
You cannot reply to topics in this forum