Distributed expert systems for queueing network capacity planning

Queueing network capacity planning can become algorithmically intractable for moderately large networks. It is, therefore, a promising application area for expert systems. However, a survey of the published literature reveals a paucity of integrated systems combining design and optimization of network-based problems. We present a distributed expert system for network capacity planning, which uses Monte Carlo simulation-based optimization methodology for queueing networks. Our architecture admits parallel simulation of multiple configurations. A knowledge-based search drives the performance optimization of the network. The search process is a randomized combination of steepest descent and branch and bound algorithms, where the generating function of new states uses qualitative reasoning, and the gradient of the objective function is estimated using a heuristic score function method. We found a random search based on the relative order of the performance gradient components to be a powerful qualitative reasoning technique. The system is implemented as a loosely coupled expert system with components written in Prolog, Simscript and C. We demonstrate the efficacy of our approach through an example from the domain of Jackson queueing networks.