The Reengineering of Relational Databases Based on Key and Data Correlations

This paper addresses the problem of recovering object-oriented schemata from relational databases. Solutions to this problem are particularly useful for designing wrappers for federated database systems. Our goal here is to describe a reverse engineering methodology for the DOK federated database system (Tari et al. 1996), enabling the wrappers to express relational schemata as object-oriented schemata which are made available for di erent DOK's services. The reverse engineering methodology we propose involves two main steps. The rst details the classi cation of relations to re ect the di erent objectoriented constructs, whereas the second step consists of applying a set of rules to generate these constructs based on the di erent information contained in local databases, both at the schema and data level. The classi cation of a relational schema consists of partitioning relations into three categories: base relations (relations which form the core classes of the target object-oriented schema), dependent relations (relations describing binary relations between classes), and nally composite relations (relations describing ternary relationships between classes). The translation of the classi ed relations into object-oriented constructs is performed by analysing them according two levels of correlation: (i) the degree of correlation between keys of the relations, (ii) and the degree of c IFIP 1997. Published by Chapman & Hall correlation between the tuples of the relations. Analysing these correlations uncovers implicit classes and relationships contained in relational schemata.