Knowledge-based monitoring and control of distributed systems
Keith A. Lantz, Bruce L. Hitson · 1989
As distributed systems and applications become more numerous and more complex, it becomes increasingly important to develop tools and techniques for studying and understanding their behavior in a systematic way. Monitoring and control (M&C) refers to the process of collecting information about a system, interpreting it, and finally using it to affect system behavior. We show that a knowledge-based approach is both necessary and appropriate for M&C of complex message-based distributed systems. Production rule systems have been used with success in many diagnostic situations, however the declarative representation style they impose does not efficiently preserve the important temporal relationships between packets. To achieve good performance, we have augmented the traditional purely declarative rule system approach with cached temporal linkage information and fast procedural analysis routines for dealing with the common operation of localized search. MC we have found that it is also appropriate in the M&C domain. However, scheduling or search control in blackboard systems is highly application-specific and must be tuned for M&C applications. We have developed multi-level weighted search control to overcome some of the scheduling problems that are encountered in domains such as M&C where shallow heuristic knowledge predominates. Our ideas have been developed and refined in the context of two implementations of monitoring and control expert systems: an automated TCP/IP network protocol analyst, and a system for both performing Time Warp distributed simulations and analyzing the results. We describe these implementation in detail, and show how their common aspects form the basis for KNOBS--a knowledge-based M&C shell and environment that is suitable and practical for studying a wide range of distributed systems.