BI Design and Architecture
Posted by thomasivarssonmalmo on April 21, 2007
The data source for a cube is normally a starschema that consist of dimension tables and one or several fact tables. I have also seen solution were you build a cube directly on an OLTP-system like Axapta or Navision but there are a number of problems with this approach.
Kimball and Inmon are the two gurus in this area of how you should build the structures behind the SSAS2005.
Inmon have this site: http://www.inmoncif.com/about/
Kimball have this site: http://www.kimballgroup.com/
My company uses Inmons approach of the corporate information factory that requires a staging database, an enterprise data warehouse(called EDW) and several data marts(called DM) on top of EDW.
In the stage database you simply import data from the source systems with changing anything. The EDW database is a normalized strucure and the Data Mart is the database for analytics.
Kimball, on the other hand, is not that keen on a large central data ware house, but insted talks about general dimension models that can be used in separate data marts. So Kimball use a stage database and several data marts on top of that.
Here is a comment from Kimball regarding Inmons approach:
In do not think that the dfference between the gurus is that big. I usually favour Kimballs more practical approach and often start my project with thinking about the dimensions. Still I follow my companies guidelines and build the data warehuse.