Abstract. Partitioning and allocation of relations is an important component of the distributed database design. Several approaches (and algorithms) have been proposed for clustering data for pattern classification and for partitioning relations in distributed databases. Most of the approaches used for classification use square-error criterion. In contrast, most of the approaches proposed for partitioning of relations are either ad hoc solutions or solutions for special cases (e.g., binary vc.r tical partitioning).In this paper, we first highlight the differences between the approaches taken for pattern classification and for distributed databases. Then an objective function for vertical partitioning of relations is derived using the square-error criterion commonly used in data clustering. The objective function derived generalizes and subsumes earlier work on vertical partitioning. Furthermore, the approach proposed in this paper is shown to be useful for comparing previously developed algorithms for vertical partitioning. The objective function has also been extended to include additional information, such as transaction types, different local and remote accessing costs and replication. Finally, we discuss the implementation of a distributed database design testbed.

The design of distributed databases is an optimization problem requiring solutions t o several interrelated problems: data fragmentation, allocation, and local optimization. Each problem can be solved with several different approaches thereby making the distributed database design a very dificult task. In this paper, we address the problem of n-ary vertical partitioning problem and derive an objective function that generalizes and subsumes earlier work. The objective function derived in this paper provides a basis for developing heuristic algorithms for vertical partitioning. The objective function is also useful for comparing previously proposed algorithms for vertical partitioning. Finally, we indicate the current status of implementation of a testbed.