On designing relational database schemes for efficient query processing

Abstract

In relational databases, operations including joins sometimes become a cause of performance problems. We study scheme transformations to reduce the number of joins used by a class of routine database activities whose access patterns are fixed. Two scheme operations, merge and copy, are defined to transform relation schemes. The former merges two relation schemes into one, and the latter copies a functional dependency of a relation scheme to its tightly related relation scheme. Necessary and sufficient conditions are shown for applying these operations to a database scheme such that no new anomalies occur. From the necessity of the conditions, the database scheme obtained by exhaustive applications of these operations as long as the conditions hold, is optimal in the sense that further application of one of the scheme operations to the database scheme would cause anomalies. It is also proved that the resulting view of a retrieval on the original database scheme is equivalent to that of the corresponding retrieval on the new database scheme.