The Schneidewind software reliability model revisited

Abstract

A software reliability model based on nonhomogeneous Poisson process {NHPP) was proposed in Schneidewind[ I ] in 1975. This model seems to be the first of this kind. Since then, many other NHPP models have been suggested and studied by various authors. Surprisingly, the paper by Schneidewind has not been widely referred to. In this paper, we show that several NHPP models can be derived based on the general assumptions made by Schneidewind. Also, we note that in the original paper, there are several interesting approaches worth further consideration. To mention a few, Schi&wind modelled both the software correction process and the software detection process. Methods of weighted least squares were adopted together with the software reliability forecasting based on previous measurements. In this paper, the Schneidewind model is revisited and some further research results are presented 1: IntroductionThe software failure process is often described by a nonhomogeneous Poisson process (NHPP). There are many papers discussing this type of models, see for example [2-91 for a general treatment of NHPP software reliability models. It seems that the first NHPP software reliability model was published by Schneidewind [l] where an NHPP is used to model the occurrence of faults during software testing.Since then many software reliability models have been proposed and studied. However, the paper by Schneidewind has not been widely referred to. In this paper, we first review the most important contributions by Schneidewind and show that many other NHPP models can be derived based on the assumptions made in fax: (46) 13 28 2742 the original paper by Schneidewind [l]. Also, we note that there are several other interesting appaches studied by Schneidewind worth further considemtion.To name some of them, Schneidewind modelled both the software fault detection process and the software fault correction process. The latter one is assumed to be a delayed fault detection process. We feel that as one of the earliest NHPP software reliability models, the Schneidewind model is an important one in the family of software reliability models. In Schneidewind [l], methods of weighted least squares are used together with some software reliability pmliction methodologies. This is also an important topic in the application of software reliability models. In this paper, the Schneidewind model is revisited and some research directions are furtber discussed. In Section 2, the essential contributions in the Schneidewind paper are reviewed and we unify some other existing software reliability models that are closely related to the Schneidewind model. In Section 3, we study some modelling approaches continuing that presented in the paper by Schneidewind, especially the modelling of fault-correction process. Finally, in Section 4 we discuss some related estimation and prediction problems such as the use of incomplete data and the methods of weighted least squares. 2: The Schneidewind model and the relation to some other modelsThe Schneidewind model is m...

Analysis of Software Release Problems Based on Fault Detection and Correction Processes

“…In order to gain a high reliability, the developers usually take a lot of time and do testing again and again. According to the above discussion, we could make the conclusion: increasing the risk cost coefficient 4 c will also lead to a longer testing time. On the other hand, different from the traditional analysis with a single fault-detection model, we use the combined fault-detection and fault-correction models discussed above.…”

“…With the limited assumption, Xie and Zhao [4] extended the time lag to be time-depended. As the detected faults become increasingly difficult to remove and correct, the time delay becomes much longer.…”

“…In Case 3, we increase the risk cost coefficient 4 c from 3,000 to 5,000 and compare the results of the optimal release time. We can find out that the optimal release time * T of Case 3 is 33.6 weeks, while that of Case 1 is 31.8 weeks.…”

“…We can find out that the optimal release time * T of Case 3 is 33.6 weeks, while that of Case 1 is 31.8 weeks. The risk cost coefficient 4 c has significant effect on the optimal release time. Compared the difference between Case 2 and Case 3, the total cost of Case 2 has more significant increase than Case 3.…”

An effort allocation model considering different budgetary constraint on fault detection process and fault correction process

“…Numerous proposed SRGMs only consider the fault detection process and are based on the assumption that a fault is removed instantly as soon as it is detected. However, this theory may not be logical as the detected faults are rarely corrected instantly (Gokhale et al, 1998;Schneidewind, 1975Schneidewind, , 2003Ohba, 1984;Xie & Zhao, 1992). Generally, whenever a fault is detected, it is reported first, then diagnosed and finally it is removed.…”