Models for Evaluating the Performability of Degradable Computing Systems.

Abstract

Recent advances in multiprocessor technology have established the need for unified methods to evaluate computing systems performance and reliability. In response to this modeling need, this dissertation considers a general modeling framework that permits the modeling, analysis and evaluation of degradable computing systems. Within this framework, several user-oriented performance variables are identified and shown to be proper generalizations of the traditional notions of system performance and reliability. Furthermore, a time-varying version of the model is developed to generalize the traditional fault-tree reliability evaluation methods of phased missions. The modeling and evaluation methods considered in this dissertation provide a relatively straightforward approach to integrate reliability and availability measures with performance measures. The hierarchical decomposition approach permits the modeling and evaluation of a computing system's subsystems (e.g., hardware, software, peripherals, interfaces, user dem and systems) as a whole rather than the traditional methods of evaluating these subsystems independently. Accordingly, it becomes possible to evaluate the performance of the system software and the reliability of the system hardware simultaneously in order to measure the effectiveness of the system design. Moreover, since the performance variables considered in this study permit the characterization of system performance according to the application needs of a system, the results obtained represent more accurate assessments of the system's ability to perform than the existing performance or reliability measures.