A New Client-server Architecture for Distributed Query Processing a New Client-server Architecture for Distributed Query Processing

This paper presents the idea of \tuple bit-vectors" for distributed query processing. Using tuple bit-vectors, a new two-way semijoin operator called 2SJ++ that enhances the semijoin with an essentially \free" backward reduction capability is proposed. We explore in detail the beneets and costs of 2SJ++ compared with other semijoin variants, and its eeect on distributed query processing performance. We then focus on one particular distributed query processing algorithm, called the \one-shot" algorithm. We modify the one-shot algorithm by using 2SJ++ and demonstrate the improvements achieved in network transmission cost compared with the original one-shot technique. We use this improvement to demonstrate that equipped with the 2SJ++ technique, one can improve the performance of distributed query processing algorithms signiicantly without adding much complexity to the algorithms. Abstract This paper presents a new client-server architecture to process distributed queries. We transform an N-way join query into an N-way join operation on the join attribute values only, and a concatenation operation to assemble the nal join result. Rather than relying on conventional semijoin techniques, we use tuple bit vectors to eeectively reduce the network overhead. We demonstrate that under this new architecture, the autonomy of the remote sites involved in executing the distributed query is preserved to a maximum degree, no direct communication link or network traac among the remote sites is necessary, the available access paths on the participating relations can be eeectively utilized and each joining relation is usually scanned only twice. The bulk of the processing is done by the client, allowing better utilization of the server. Several techniques are proposed to handle the potential processing bottlenecks that might occur before and during the nal assembling stage. Finally we explore some of the query optimization and performance issues encountered when following this architecture to process distributed N-way join queries.