Studies on Reliability Analysis for Microprocessor Systems

Abstract

This thesis treats several stochastic models of μP systems. Using the theory of Markov renewal processes , the reliability measures such as the mean times to system failure and to completion of the process are obtained. Moreover, the expected costs are derived and optimal policies which minimize them are analytically discussed. Finally, numerical examples of each model are given and some useful discussions are made . This thesis is divided into 9 chapters. Chapter 1 states fault tolerant techniques and microprocessors (μPs). Chapter 2 considers a μP system with a watchdog timer (WDT) which is preventively maintained at time T and at reset number N. Next, Chapter 3 treats a system where a main processor (MPu) has N watchdog processors (WDPs) with self-checking. To prevent that the MPu becomes faulty, the stochastic model to determine the number of WDPs is formulated. The μP unit which consists of μP and WDP has been recently used. Chapter 4 and Chapter 5 study a system with N μP units. It is assumed in Chapter 4 that a μP is in faulty state if more than K resets have occurred at time T. From the viewpoint of real-time processing of the system, it would be necessary to have the function which completes one processing within a certain limit time. It is assumed in Chapter 5 that a μP is in faulty state if it does not finish one processing until a limit time T. Chapter 6 considers a system with N TMR (Triple Modular Redundancy) units in which each unit consists of μP and WDP. Introducing the concept of complexity, an optimal number of TMR units which minimizes the expected cost is discussed. Chapter 7 deals with the problem for improving the reliability of a μP system with network processing. An optimal policy which minimizes the expected cost until a network processing is successful is discussed. Further, Chapter 8 considers the reliability problem of a μP system whose errors can be detected by using signatures. An optimal division number of a job is discussed. Finally, Chapter 9 summarizes the results derived in this thesis.