Parallel Evaluation of Relational Queries on a Network of Workstations

In this paper we propose an innovative approach to handle ``read-most'' data bases. This approach is based on a parallel extension, called parallel relational query evaluator, working over a network of workstations, in a coupled mode with a sequential Database Management System (DBMS). We present a detailed architecture of the parallel query evaluator and focus on the management of data during executions and transmissions, especially through macro-pipelining. We then present Enkidu, the prototype that as been build according to our concepts. We finally expose a set of measurements, conducted over Enkidu, highlighting both the specific performances of macro-pipelining and the global ones of Enkidu.