TOWARDS A VIABLE METHODOLOGY FOR LOGICAL RELATIONAL DATABASE DESIGN

Abstract

Traditionally, logical relational database design has been accomplished using two differing paradigms. The normalization framework starts out with a universal relation schema and transforms this into a collection of relation schemas in good normal form. A second framework involves the use of semantic data models such as the Entity Relationship (ER) model. In this second approach, an ER schema captures the data requirements o[ an enterprise, and this is then transformed into relation schemas in good normal form by (i) mapping it directly to relation schemas and normalizing these, or (ii) transforming it to some ER normal form which can then be translated directly to relation schemas in good normal form. In this thesis, we demonstrate that the above approaches to logical relational database design are inadequate. Our objections can be summarized into two points. First, relation schemas in classical normal forms do not adequately characterize a database schema that is 'good', as measured by the criteria proposed by Beeri et al. [9]. Second, the design frameworks themselves suffer from a number of anomalies. On one hand, the normalization framework is shown to be unrealistic because it assumes that an enterprise's data requirements can be adequately captured in a universal relation schema satisfying the Universal Relation Assumptions. On the other hand, the ER-based design methods, while providing a more intuitive means for documenting real world data requirements, often result in erroneous design or are themselves severely restrictive models due to the semantic mismatch between the ER and relational formalisms.