Imperfect debugging models with fault introduction rate for software reliability assessment

TLDR

Software reliability models typically assume perfect fault removal, which is unrealistic. The paper proposes two software reliability models that account for imperfect debugging, where correcting faults can introduce new ones. Both models use a nonhomogeneous Poisson process with a fault detection rate proportional to the sum of remaining and introduced faults, and derive quantitative reliability measures and maximum likelihood estimators. Numerical examples demonstrate the application of the two models to software reliability analysis.

Abstract

In general it is considered to be unrealistic in software reliability modelling to assume that the faults detected by software testing are perfectly removed without introducing new faults. In this paper we propose two software reliability assessment models with imperfect debugging by assuming that new faults are sometimes introduced when the faults originally latent in a software system are corrected and removed during the testing phase. It is assumed that the fault detection rate is proportional to the sum of the numbers of faults remaining originally in the system and faults introduced by imperfect debugging. These two models are described by a nonhomogeneous Poisson process. Several quantitative measures for reliability assessment are derived, and the maximum likelihood estimations of unknown model parameters are presented. Finally, numerical examples of software reliability analysis based on these two models are shown.