A knowledge-based approach to automated interpretation of performance model results

Abstract

Models used to evaluate the performance of complex time-critical computer systems can produce voluminous amounts of data. The paper discusses the integration of knowledge-based systems with performance models to produce knowledge-based performance evaluation systems that provide automated interpretation of model results. A three stage conceptual model of interpretation is developed. Design alternatives for developing knowledge-based performance evaluation systems are explored. The critical dependencies among automated interpretation, design representation, and performance models are described, and a threshold-based strategy for problem identification using the notion of causal factoring is discussed. A prototype system called PEDAS illustrates the authors approaches.<>