In data warehousing there are two commonly acknowledged approaches to building a decision support infrastructure, and both can be implemented using the tools available in SQL Server 2005 with Analysis Services 2005. It is worth understanding these two approaches and the often-cited difference of views that result. These views are expressed most overtly in two seminal works: The Data Warehouse Lifecycle Toolkit by Ralph Kimball, Laura Reeves, Margy Ross, and Warren Thornthwaite, and Corporate Information Factory by Bill Inmon, Claudia Imhoff, and Ryan Sousa.
Kimball identified early on the problem of the stovepipe. A stovepipe is what you get when several independent systems in the enterprise go about identifying and storing data in different ways. Trying to connect these systems or use their data in a warehouse results in something resembling a Rube-Goldberg device. To address this problem, Kimball advocates the use of conformed dimensions. Conformed refers to the idea that dimensions of interest — sales, for example — should have the same attributes and rollups (covered in the "Aggregations" section earlier in this chapter) in one data mart as another. Or at least one should be a subset of the other. In this way, a warehouse can be formed from data marts. The real gist of Kimball's approach is that the data warehouse contains dimensional databases for ease of analysis and that the user queries the warehouse directly.
The Inmon approach has the warehouse laid out in third normal form (not dimensional) and the users query data marts, not the warehouse. In this approach the data marts are dimensional in nature. However, they may or may not have conformed dimensions in the sense Kimball talks about.
Happily it is not necessary to become a card-carrying member of either school of thought in order to do work in this field. In fact, this book is not strictly aligned to either approach. What you will find as you work through this book is that by using the product in the ways in which it was meant to be used and are shown here, certain best practices and effective methodologies will naturally emerge.