The Study Of Software Reliability Model Considering Imperf Fault Correction Process

With the widespread used of software, the requirement about software quality is more and more rigorous. And the software reliability is the most important aspect of software quality. There are many factors affect the research on software reliability, mainly include four factors: the inherent characteristics of software, test conditions, software failure data and human interventions. Therefore, the researchers have to choose, simplify and restrict some various factors as to establish the reliability of the mathematical model and the basis of expected model. Base on different assumptions and starting point of research, people had developed many different software reliability prediction models: include J-M model, S-W model, Nelson model, Littlewood-Verrall model, Musa model, NHPP model and so on. However, in so many models, no existing model has been proved universally applicable, because the assumptions are the mainly basis of the establishing models. And there are many differences between the ideal assumptions of existing models and the reality software development process.The classical models (include J-M model and G-O model), which were based on ideal fault correction process, ignore the effect of the realistic fault correction process. In fact, the realistic fault correction process is not perfect. The fault can not be corrected immediately after it be detected, and the fault introduction and removal efficiency should be considered. The fault removal rate is not constant, but depends on time. According to the reality software development experience, the fault removal rate is decrease with the test time increase. Also the test environment is different from operation environment. In this paper, considering the actual impact of fault correction process, some new assumptions are proposed ,and a new improved model based on the software reliability model which was proposed by Xuemei Zhang is created. Some simulation experiments are carried out in the new model and selected ones with two set of software failures data which were openly published. Through analyze and compare one of the improved model, it ultimately proves the superiority of the improved prediction model, and the new models is more suitable to be used in actual software reliability process.