Software reliability growth models incorporating imperfect debugging with introduced faults

In order to quantitatively assess software reliability, various software reliability growth models have been proposed. These models describe software failure occurrence and software fault detection in dynamic environments, such as testing in the software development process or in operational use. In most of these models, perfect debugging is assumed, where the fault which caused the software failure is corrected accurately and completely. In practice, however, this is not always true, since many software engineers have the experience that a new fault is introduced through the correction process. In other words, fault correction is an imperfect debugging process. From such a viewpoint, this paper proposes software reliability growth models for the imperfect debugging environment, in which the model is made more practical by considering the possibility of introducing a new fault in the fault correction process. It is assumed that there exist two kinds of software failures generated in the dynamic environment of the software system. One is software failure due to an inherent fault which latently exists before operation, and the other is a software fault which is randomly introduced during operation. The software failure occurrence is described by a nonhomogeneous Poisson process. Quantitative measures for software reliability assessment are derived. The application of the models to the actually observed data is shown, and the goodness of fit of the models is compared to that of conventional models. (C) 1998 Scripta Technica.