The Application of Functional Dependency Theory to Relational Databases

This paper examines three areas where the application of functional dependency theory to relational databases has had an impact. These areas are relational views, database translation and logical database design. The paper refines our earlier work on relational views in which the concepts of consistent and updatable views were proposed. In the area of database translation, it identifies six levels of information preserving relational transformations. Finally, a relational database design algorithm is proposed which combines the best features of the classical synthesis and decomposition approaches while avoiding their identified shortcomings. INTRODUCTION This paper shows, that functional dependency theory provides a unified framework for the analysis of several database problems. The paper represents both a synthesis and extension of our earlier work on each of these problems. The first section contains background infor- mation to familiarize the reader with the terminology and concepts of functional dependency theory. In the second section, functional dependencies are used to define the concepts of consistent and updatable views. Basically, a 'consistent view' is one whose interpretation is consistent with that of the underlying database. An 'updatable view' is one whose updates can be translated into appropriate updates of the underlying database. Together, these definitions describe a class of useable views. In section three, six levels of information preserving relational transformations are identified. These levels reflect the different degrees with which transformations preserve the update and retrieval properties of databases. These levels are important since previously many applications have mistakenly assumed that all the information embodied in their database is preserved by the restructuring operations they have employed. Finally, some shortcomings of the classical synthesis and decomposition approaches to relationship database design are identified in the fourth section. A new algorithm is then proposed which combines the best features of these approaches while avoiding their iden- tified shortcomings.