A multiprocessor organization for very large relational database

Abstract

The authors present a decentralized architecture for a multiprocessor database machine which supports multiple-user, online, very large relational database systems. A corresponding processor allocation strategy is proposed which applies the dataflow technique to distribute processors across all nodes of the vertically layered query tree and to pipeline pages of intermediate relation between them. Simulations show that the query execution time is 20%-25% less than that in the DIRECT database machine. The time used to process the messages in the authors' database machine is almost 40% less than the time used in DIRECT. The simulation also indicated that the machine's structure is less sensitive to the I/O bound than the DIRECT machine.<>