Architectures for parallel query processing on networks of workstations

Networks of workstations (NOWs) are cost-effective alternatives to multiprocessor systems. Recently, NOWs have been proposed for parallel query processing. This paper explores the various architectures for using NOWs for parallel query processing. We start our discussion with two basic architectures: centralized and distributed. We identify the advantages and disadvantages associated with these architectures. We then propose a hierarchical architecture that inherits the merits of the centralized and distributed architectures while minimizing the associated pitfalls. We also discuss the performance and research issues that require attention in all three architectures.