Building the Data Warehouse

Скачать в pdf «Building the Data Warehouse»

From a short-term perspective, there is some merit to this argument. But from a long-term perspective, a data mart is never a substitute for a data warehouse. Figure 3.56 shows why.

The structure of the data found in the data mart is shaped by the particular requirements of the department. The finance department will have one structure for its data mart, the sales department will have another structure for its data mart, and the marketing department will have another data structure for its data mart. All of their structures will be fed from the granular data found in the data warehouse.

The data structure found in any given data mart is different from the data structure for any other data mart. For example, the sales data mart data structure will be different from the marketing data mart data. The data mart structures are typically known as star joins and contain fact tables and dimensions. The data mart structures are typically known as multidimensional structures and are served by OLAP technology.

Because there is a different data structure for each data mart, making any data mart into a data warehouse doesn’t make sense. When a data mart star join is made into a data warehouse, the data warehouse is optimal for one data mart and its users and is not optimal (or really usable) for anyone else. Data marts produce structures that are not reusable except by someone operating in the department that is optimized.

Data mart data structures—in general, across the enterprise—are not reusable, are not flexible, are not useful as a foundation for reconciliation, and are not standing ready for a new set of unknown requirements. But the normalized granular data found in a data warehouse is indeed all of those things.

Скачать в pdf «Building the Data Warehouse»