Principles of Static Clustering for Object Oriented Databases

This thesis addresses problems of clustering in the context of object oriented databases, from a theoretical and a practical point of view. We identified and formulated the clustering problems of Object Oriented Systems, derived optimal solutions to those problems, proposed practical clustering algorithms, and completed a thorough performance evaluation of those algorithms as well as many others. Our major results are the following: (1) With respect to the clustering problem, object access patterns can be reasonably modeled using stochastic access models. (2) The fundamental Optimal Clustering problem is NP-Complete, being an instance of hyper-graph partitioning. Asymptotically, however, optimal clustering reduces to probability ranking partitioning and has low complexity (N log N). (3) We proposed two new heuristic clustering algorithms that outperform most of the existing clustering techniques if access patterns information is available. However, good static clustering mappings are expensive to obtain and can be very sensitive to the access patterns.