9 - Celko on data warehouses: Techniques, successes, and mistakes

Abstract

A data warehouse should be as large as possible. A data warehouse is designed for a few trained users who engage in long, complicated sessions. These sessions are ad hoc queries or statistical analyses. A data warehouse should not be normalized because the data never changes while it is in use. The data warehouse queries deal with aggregates rather than with single transactions. Computing the same aggregations over and over is a waste of time. But storage is cheap, so it is possible to store the aggregations in the data warehouse. Hybrid OLAP (HOLAP) allows one to construct aggregations in a relational system and store them in a multiple dimensional databases (MDDB). Setting up the data warehouse helps companies do data audits and clean up their production databases. The second effect of a data warehouse project is that people look at the entire enterprise as an integrated process, instead of separate parts only vaguely related to each other. The third effect of a data warehouse project is that one gets useful information out of the warehouse.