An Expert System for Performance

The rapid growth in size and complexity of distributed systems and the use of heterogeneous components require eeective tools to manage their resources. Expert system technology possesses the potential to manage such large and complex systems based on heuristic knowledge in near real time. This paper presents a general approach to designing an expert system for the management of performance of a distributed function which is built based on the services provided by several application servers distributed across the network. Furthermore, the expert system re-allocates the application servers so that all the clients (who are running the same distributed function from diierent sites) will experience similar function response times. We also present an implementation of this system using a commercial expert system shell and inference engine. This presented expert system design has three components: (1) resource monitor, (2) performance optimizer, and (3) the Management Information Base (MIB). We demonstrate through an example how this expert system can be used to optimize the distribution of application servers of a client-server based distributed system with three ethernet-based networks interconnected by an FDDI backbone network.