Infinite failure models for a finite world: A simulation study of the fault discovery process

Many software reliability growth models have been published in journals and conference proceedings over the last 20-25 years. Each one has been justified based on theoretical or empirical evidence. Although there are many ways of classifying these models, a particularly interesting classification involves distinguishing models based upon the asymptotic expected number of total failures. These models are termed infinite and finite failure models since each one expects infinite and finite failures respectively as time approaches infinity. As might be expected, theoretic and especially empirical justification for the appropriateness of infinite failure models came after justifications for finite failure models. Infinite failure models were associated with weak fault repair systems or possibly with highly nonuniform usage, although this term is non-specific. We demonstrate through simulations of black-box testing (and/or field usage) that infinite failure models are appropriate in situations where there is perfect and near-perfect repair and where usage is uniform for the vast majority of the system