Use of importance sampling and related techniques to measure very high reliability software

Computer-based control systems have grown more complex over the past two decades. Thus, the software aspects of system reliability are an increasingly important concern. Current methods of software and system reliability prediction-whether measurement based or incorporating reliability growth models-cannot accurately predict failure rates of greater than 10^-6 per mission hour. This paper describes a new methodology for more accurately predicting failure rates of very high reliability systems. The methodology enhances conventional measurement-based reliability assessment with a method incorporating the results of stress testing called importance sampling. By means of importance sampling in conjunction with a system model, acceleration factors can be associated with stress testing much as is currently done with elevated temperature life testing of hardware components