MODELLING APPROACH TO THE EVALUATION OF COMPUTER SYSTEM PERFORMANCE.

Abstract

Two different modelling techniques, regression and simulation modelling, have been applied and a method developed of combining their use within a multilevel hybrid modelling framework. The main objective is to demonstrate the feasibility and value of this approach to the modelling and evaluation of computer system performance. The approach has been demonstrated by modelling a CDC 6000 computer system at three levels of detail. At each level, a self-contained model of the system has been developed. The Workload Model is a purely regression model of computer system performance. It was developed after a performance analysis and analysis of residuals. The model expresses a batch job elapsed time as a function of job resource demands and the load on the system. The model has been successfully validated. The Load Adjusting Model is a hybrid simulation/regression model in which a simulation framework is created which models job arrival and termination. Within this framework, regression techniques are used. The Memory Management Model is developed from the Load Adjusting Model by a systematic expansion of detail. The Memory Management subsystem is simulated in considerable detail, whereas the rest of the system is modelled in much less detail. Both hybrid models have been calibrated and validated.