by Angela Guess
Jonathan Geiger recently tackled the question of why it takes such a long time to develop “a data model that can be used for designing the data structures for a data warehouse or data mart. To answer this question,” Geiger writes, “we should consider the major steps involved in developing the data model and how long each step should take. For simplicity, we’ll use the following steps, recognizing that these are high-level groupings and that the specific steps may vary based on the governing methodology and architecture: requirements and business rules gathering; source analysis; logical model development; and physical model development.”
Geiger goes into further detail on each step, noting that the first two take the most time: “Before data modeling work can begin, the data modeler needs to understand requirements from three different aspects. On the surface, the primary driver may appear to be a determination of the data that is needed, but that only scratches the surface. The needed data elements determine the attributes (columns) that must be included in the data model and their placement within a normalized model. They do not, however, contribute to the overall structure.”
He goes on, “The dimensional model is structured to respond to questions for various business subjects and provide a means of navigating through the data for the desired level of summary and filtration. A key factor in designing these models is understanding the requirements in terms of the business questions that need to be answered. Those questions can be captured using various techniques (e.g., fact-qualifier matrix) to help determine the number of star schemas that are needed and the questions to be answered by each. The third requirement that helps the data modeler is the reason the data is needed. In other words, ‘Why do these business questions need to be answered?’ That requirement provides the modeler with information to anticipate future needs so that a more extensible model can be built.”
