Computer Science Department University of Maryland College Park
Recent developments in software and hardware changed the way database systems are built and operate. In this paper we present database architectures based on the Client-Server paradigm and study their performance and scalability under different query/update workloads. The architectures are: Standard Client-Server, Client-Server with Multiple Disks, and Enhanced Client-Server. Data replication and client query result caching are used as the main mechanisms to improve the query throughput. The role of the server is to maintain system-wide data consistancy and in the case of Enhanced Client-Server to selectively propagate updates on demand. Our study shows that except for the case of mostly update workloads, the Standard Client-Server architecture is outperformed by the two other architectures by one or more orders of magnitude. The Client-Server with Multiple Disks architecture offers performance comparable to that achieved by the Enhanced Client-Server for up to 100 clients, but the latter scales up a lot better for higher numbers of clients.