A Hybrid Object Clustering Strategy for Large Knowledge-Based Systems

Object bases underlying knowledge-based applications tend to be complex and require management. This research aims at improving the performance of object bases underlying a class of large knowledge-based systems that utilize object-oriented technology to engineer the knowledge base. In this paper, a hybrid clustering strategy that beneecially combines semantic clustering and iterative graph-paritioning techniques has been developed and evaluated for use in knowledge bases storing information in the form of object graphs. It is demonstrated via experimentation that such a technique is useful and feasible in realistic object bases. A semantic speciication mechanism similar to placement trees has been developed for specifying the clustering. The workload and the nature of object graphs in knowledge bases diier sig-niicantly from those present in conventional object-oriented databases. Therefore, the evaluation has been performed by building a new benchmark called the Granularity Benchmark. A segmented storage scheme for the knowledge base using large object storage mechanisms of existing storage managers is also examined.