The effects of database normalization on decision support system performance

Relational database normalization strives to minimize update anomalies and data redundancy, often at the cost of performance. Transactional systems typically require a higher degree of normalization since data are updated more frequently than in read-intensive decision support systems. While these reasons for and effects of normalization can be considered common knowledge, there are hardly any empirical studies on the query performance implications of various degrees of normalization in decision support systems. That is, it seems that the magnitude of the effects of normalization is not widely understood, even though performance implications are of importance to managers and analysts utilizing decision analytics, and for end-user information needs to be timely satisfied. In this study, the effects of normalization on a decision support database were tested for three popular SQL/relational database servers. The results raise serious concerns about the conventional consensus on the performance gains incurred by the reduced number of table joins. Even for small-sized databases, the penalties due to the extra volume caused by redundancy associated with lower normal forms seem larger than the performance gains due to the reduced number of joins. These results have practical implications on which design principles should be followed for efficient decision support system databases.