Cooperative Transaction Processing between Clients and Servers

Business rules are often implemented as stored procedures in a database server. These procedures are triggered by various clients, but the execution load is fully centralized on the server. We improve the overall response time and increase server throughput by balancing this load between the server and the clients. In our novel scheme, parts of the stored procedures are executed on cached data at the client. The critical issue in such a system is the trade-off between synchronization effort among clients and the server and the increase in systems performance gained by load balancing. We present an architecture using an optimistic synchronization protocol together with an algorithm for data verification at the server. The experience we gained through a detailed case study, based on a real-life eCommerce application, shows that in many situations a considerable speed-up is possible.