Evaluating the Performance of Software-Reliability Models

In this paper we investigate a method of quantifying the performance of software-reliability models that was proposed by Littlewood. The method is essentially a Kolmogomv-Smirnov goodness-of-fit test. We demonstrate that although the Littlewood method is rather unconventional, the standard Kolmogorov-Smirnov tables (for the simple hypothesis) can be directly applied. On this basis, we show the kind of performance that can be anticipated from an ideal software-reliability model. We also present a new performance measure which overcomes a weakness in the Littlewood method