A decomposition method for reliability analysis of real-time computing systems

Abstract

A functional decomposition method for reliability analysis, based on the specific characteristics of real-time systems, is proposed. The system under scrutiny is decomposed into hardware modules and software tasks. The hardware structure and fault occurrence process are captured by a continuous-time, discrete-state Markov model. The software structure and fault/error handling process are described by several discrete-time, discrete-state Markov chains. Eventually, a reduced Markov model, which accounts for system hardware and software structure and fault occurrence and fault/error handling processes, is generated. Closed form solutions are derived for time dependent and steady state probabilities, taking into consideration the weights of permanent and transient fault classes. Failure rates and coverage probabilities are the input parameters of the model. Failure rates are provided by reliability handbooks or circuit life testing. Fault injection is usually used for deriving coverage probabilities. A fault injection simulation algorithm, based on functional decomposition, is described and the achievable simulation time savings are evaluated. The main advantages of the proposed modeling approach are: functional decomposition closely follows the hardware and software structure of the real system. This simplifies both the evaluation of the model input parameters and model validation; closed form solutions overcomes numerical problems, e.g. largeness and stiffness, typically encountered when Markov models are employed for reliability analysis; and considerable savings, in terns of simulation time, are achieved. >