Markovian software reliability measurement with a geometrically decreasing perfect debugging rate

The assumption of perfect debugging is a controversial issue in software reliability modeling. Most existing software reliability models assume that all faults causing software failures are detectable and correctable, and that no new faults are introduced into the software system by debugging activities. This paper discusses a software reliability model for an imperfect debugging environment where the detected faults are not always corrected and removed from the system. Taking notice of the cumulative number of corrected faults, we relate the uncertainty of debugging activities to the increase in complexity of corrected faults. The stochastic behavior of the fault-correction phenomenon with imperfect debugging is described by a Markov process. Several quantitative measures for software reliability assessment are derived from this model.