YSCOPE: A Shell for Building Expert Systems for Solving Computer-Performance Problems

Abstract

Solving computer-performance problems could be greatly simplified by using an expert system which detects when a performance problem exists, diagnoses its cause, recommends system changes to eliminate the problem, and explains why particular recommendations were or were not made. Currently, such computer-performance expert systems have a knowledge base of rules such as: "If MVS paging delays are excessive, then recommend removing frequently accessed user data sets from paging devices". Unfortunately, developing the knowledge base is a formidable task, and existing expert-system-building tools have many shortcomings when applied to computer-performance expert systems. Herein, we describe the Yorktown Shell for Computer Performance Experts (YSCOPE) which is a special-purpose shell for building computer-performance expert systems. YSCOPE is based on the observation that solving computerperformance problems requires two types of knowledge: knowledge of the computer system (e.g., in MVS, user data sets may reside on paging devices) and insights from queueing theory (e.g., delays increase with utilizations). By being a specialpurpose shell and by incorporating a knowledge of queueing theory, YSCOPE provides the following features needed by computer-performance expert systems but which are not provided by existing expert-system-building tools: • Facilities for detection, diagnosis, and recommendations • Explaining recommendations in terms of resource contention • Checking for errors and omissions in the knowledge base • Potential simplification of the knowledge base since it does not incorporate queueing knowledge A prototype version of YSCOPE has been implemented in VM/PROLOG, and a knowledge base has been developed to handle paging problems in MVS using data available from RMF.