Performance Evaluation of Large-Scale Parallel Clustering in NOW Environments

This paper presents the results of a performance study of parallel data clustering on Network of Workstations (NOW) platforms. The clustering program, P-CLUSTER, is based on the mean square-error clustering algorithm and is applied to the problem of image segmentation. The parallel implementation uses a client-server model, in which the clustering task is divided among a set of clients that report their intermediate results to a single server process. Results of experiments on four NOW platforms are presented, illustrating the eeects on performance of diierent processors, network architectures, communication packages, and latency hiding techniques.