Client/server architecture in the ADAMS parallel object-oriented database system

Abstract

This paper describes issues encountered in the design and implementation of a parallel object-oriented database system. In particular, we find that the design of a client/server interface (that is, whether to use a page server or query server architecture) depends greatly on the expected application environment. We believe that the query server model is more appropriate for most scientific database applications. This paper develops the reasons for this assertion. Then we discuss the implementation of a working parallel object-oriented, database system, called ADAMS, that has been developed at the University of Virginia. ADAMS is best known as the database system underlying the “Oracle of Bacon” website, which is currently responding to over 5 million queries per year. We show that set operations such as union and intersection can be performed in a completely data-parallel fashion, and that implicit join queries can be performed with very few inter-processor messages in a way that scales well. All data access in the system can be completely monitored, so these observations are supported by experimental results. Finally, we compare our approach with an equivalent page server version.