Selecting a software reliability model based upon failure data characteristics

Classical reliability theory is fundamentally applicable and useful in the software environment, notwithstanding the observation that software does not de teriorate with age. Several models have been proposed that attempt to better characterize the failure behavior of software. Each software reliability model has advantages and disadvantages in terms of analytical complexity, predictive validity, required data, ease of understanding, and ease of use. Nine software reliability models are presented. The assumptions underlying each of the models are. presented and analyzed to determine their accuracy in modeling the behavior of software. The mathematical expressions that are used to compute reliability measures for each of the models are derived from the assumptions that form the basis of the respective models. These mathematical expressions contain parameters that are required for the computation of the various reliability measures. The methods used to estimate such parameters in the implementations of each of the models are given. The models presented are compared for reported failure data sets chosen to be representative of failure data distributions for software systems of high, medium, and low quality. Based upon these comparisons, specific models are recommended for predicting reliability measures for software systems of high, medium, and low quality, respectively.