Considering Testing-Coverage and Fault Removal Efficiency Subject to the Random Field Environments with Imperfect Debugging in Software Reliability Assessment

As the quantitative software reliability assessment means, software reliability model has become one of the most successful and practical software reliability engineering contributions. Although many models have been developed in the past 30 years, most of them take the following assumptions: 1) the software's final field environment is the same as its testing environment, so the software reliability model used in testing environment will be suitable for its field applications, 2) fault removal efficiency equals to 1, i.e. the faults detected will be removed perfectly and not part of the faults will remain. But these two assumptions do not always stand in practice. Firstly, software is usually tested in a given controlled environment, but it might be used in different field environments by different users, so due to the randomness of the field environments, the software's performance and reliability may be considerably influenced in an unpredictable way. Secondly, the fault removal efficiency is usually imperfect, i.e. the faults causing the observed failures may not be removed completely and the original faults may remain and new faults may be introduced during the debugging process. This paper aims to incorporate fault removal efficiency, testing coverage and the randomness of field environments into software reliability assessment and develop a new software reliability model. We compare the performance of the proposed model with several existing nonhomogeneous Poisson process (NHPP) software reliability growth models (SRGMs) using software failure data collected from real applications. All results show that the new model can give an improved goodness-of-fit.