Clustering relations into abstract ER schemas for database reverse engineering

Database reverse engineering (DBRE) methods recover conceptual data models from physical databases. The bottom-up nature of these methods imposes two major limitations. First, they do not provide an initial high-level abstract schema suitable for use as a basis for reasoning about the application domain: a single detailed schema is only produced at the very end of the project. Second, they provide no support for a divide-and-conquer approach: the entire database schema must be analysed and processed as a unit. This paper presents a simple solution to overcome both limitations. In our proposal, relations are grouped based on their primary keys. Each group can be perceived in two ways: as a relational schema that can be reversed engineered as a standalone DBRE project; and as an element, either an entity or a relationship, of a high-level abstract schema that provides initial insight about the application domain. We also present examples from actual large database systems.