Parallel processing and data driven implementation of a relational data base system

Abstract

A simple query to a large data base may take a prohibitively long time to process. To decrease this time in applications requiring fast response, parallel processing is necessary. Within the scope of a Relational Algebra query system, the potential parallelism in query execution is investigated. Three levels of parallelism are identified: independent parallelism, pipelining and node splitting. It is shown that, under certain conditions, the exploitation of these levels makes highly parallel processing and short response time possible. A data driven computation scheme which can implement the three levels of parallelism in a unified way is described.