This research focuses on performance issues in rule-based knowledge bases that have to manage large amounts of rules and data. An effective implementation of such systems, particularly if they are distributed, has proven to be a major stumbling block in the way of their efficient performance. In this project, parallel processing for solving the performance problem is investigated. The goals of the research are: (1) to establish a theory of parallel and distributed rule processing; and (2) based on the theory, to devise processing strategies and criteria for comparing them, so that an optimal strategy can be selected for a given rule-based program, data structures, and a parallel/distributed architecture. The results of the research will be experimentally evaluated in Net-mate, a system for management of very large communication networks (hundreds of thousands of interconnected computers), that is currently under development at Columbia University. An important issue in network management is automatic detection of fault conditions. Rule-based programming is ideal for this purpose, but it requires the analysis of very large amounts of statistical and configuration data that is distributed, leading to the type of performance problems that this research aims to solve. Hence, the Net-mate system is an ideal testbed for this research, and both projects are expected to benefit from the cooperation and provide results applicable to distributed and parallel computing.
