Investigation on the Side Effects of Denormalizing Corporate Databases

Corporate databases are usually denormalized, due to the data modelers' impetuous belief that denormalization could improve system performance. By providing a logical insight into denormalization, this paper attempts to prevent every database modeler from falling into the denormalization pit. We indicate loopholes in the denormalization advocates' assertions, and then present four criteria to analyze the usefulness and validity of denormalization; 1) the level of concurrency among transactions, 2) the database independence of the application program, 3) the independence between the logical design and the physical one, and 4) the overhead cost to maintain database integrity under various query patterns. This paper also includes experimental results to evaluate performance of denormalized and fully normalized structures under various workloads.