Data warehouse architecture
Goldfarelli & Rizzi (2009) describe two classifications of a data warehouse architecture. The first classification describes the physical separation of the data warehouse's architectural layers. The second classification is based on the methods devised by Inmon and Kimball. Inmon advocates the use of hub-and-spoke architecture, while Kimball prefers an architecture based on the data mart bus (Ariyachandra & Watson, 2006).
The advantage of the hub-and-spoke approach is data integration versus the ease of development and a lower cost of the federated approach (Hwang & Cappel, 2002).
Data warehouses are implemented in a layered architecture. In the single layer, both the OLTP and OLAP compete for system resources causing performance degradation (Goldfarelli & Rizzi, 2009). The multi-layers architectures like the two and three layer designs create a physical separation so that the two systems don't interfere with each other (Goldfarelli & Rizzi, 2009).
The three layer architecture includes a reconciled layer where a database contains data that has been cleansed, adjusted, or enhanced to provide reliable and consistent data that can be imported an mapped into the data warehouse schemata for data analytics (Goldfarelli & Rizzi, 2009).
Data warehouse development approaches
There are two major approaches to choose from when implementing a data warehouse. The first is the top-down approach, which is the most expensive, requires that the global business needs are thoroughly analyzed so that a enterprise wide data warehouse is implemented (Goldfarelli & Rizzi, 2009).
The second method, which is more universally accepted, is the less costly bottom-up approach that uses departmental facts and an incremental method of discovery to create several data marts (Goldfarelli & Rizzi, 2009).
Vowler (2002) suggests that in order to decide on which approach to use, the designer must ask where the urgent need lies. If the business need is at the highest level and the CEO can wait, then the top-down approach should be chosen. However, if the need is at the lowest levels, then the bottom-up approach should be used for a quick return on investment (Vowler, 2002).
Pep Boys implemented their data warehouse by using the bottom-up approach. They included sales people and field managers during the requirement stage (King, 2003).
Building data warehouses is an expensive undertaking. Hwang & Cappel (2002) indicate that the cost difference between hub-and-spoke and federated approaches are negligible. The average cost for hub-and-spoke and federated data marts are $4.7 million and $5.1 million respectively (Hwang & Cappel, 2002).
Business Dimension Lifecycle
When starting a data warehouse project, the designers must focus on the types of questions the users will ask (Hammergren & Simon, 2009). Business users query data that is organized in a multidimensional format where the first dimension are the columns headings and the second...