We present a new reliability model for hard real-time systems. This model uses a generic high-level formalism based upon a Markov chain with a lattice structure which represents the progress of a computation, allowing both functional and time correctness of the system to be modelled. This is an improvement on traditional system reliability models which typically focus on functional correctness, and do not adequately model the temporal properties of such systems. We provide an example of the application of this model to a recovery block system, and show that a number of important metrics may readily be derived from these results. We note an unusual feature of the failure profile data, from which we hope to derive measures of the independence of the alternates in a recovery block system.

The reliability modelling and analysis of real-time, fault-tolerant, embedded systems is considered. It is shown that many existing reliability modelling techniques are inadequate for this task, since they model only the overall system reliability, whilst the timing properties of the system are either neglected, or reduced to simple metrics. A new reliability model is derived, which permits the modelling of both overall system reliability, and the probability distribution of system completion and failure times. This model is based on a set of high level system attributes, which it is expected may beestimated from experimental data. The model is applied to the study of recovery block systems, and it is shown that the results obtained are compatible with, and extend, a number of other system reliability models. The thesis concludes with a discussion of the application of more detailed timing information to the scheduling of safety-critical real-time systems. It is shown that the additional timing information available with models such as that developed herein, allows designers to make more informed choices regarding the tradeo between safety and performance. CONTENTS

We present a new reliability model for hard real-time systems. This model uses a generic high-level formalism based upon a Markov chain with a lattice structure which represents the progress of a computation, allowing both functional and time correctness of the system to be modelled. This is an improvement on traditional system reliability models which typically focus on functional correctness, and do not adequately model the temporal properties of such systems. We provide an example of the application of this model to a recovery block system, and show thatanumber of important metrics may readily be derived from these results. We note an unusual feature of the failure pro le data, from which we hope to derive measures of the independence of the alternates in a recovery block system.