| As the functional requirements of today's software applications become more crucial, the modern commercial software products' size becomes larger. The large size and increasing complexity of the program require more development time and effort. While in contrast, its decreasing average life expectancy and early market arrival before competitors, more than ever, suggest less time and resources for its use. However, the importance of system failure on both users and developers has become more critical and wider. Thus, the impact on software reliability today, not to mention safety-critical applications, should be more emphasized.; In this dissertation we first develop a testing methodology, which improves the testing effectiveness by reducing the testing effort while simultaneously, ensuring competitive quality of software product. To this end, a new systematic procedure incorporating the characteristics of software testing into a continuous sampling plan is proposed.; When we consider both imperfect and delayed debugging, more focus has been given to developing a generalized non-homogeneous Poisson Process (NHPP) software reliability model, which reflects more realistic conditions on the software testing and debugging activities. The occurrence of a debugging event with a constant delay-interval was formulated based on the existing software reliability models that consider various aspects of testing and debugging. The Quasi-renewal assumption is then applied to estimate the delay-interval by modifying the previous constant delay-interval assumption. We finally incorporate the imperfect debugging effect into the generalized delay debugging software reliability model. Numerical examples with real failure data are presented to illustrate descriptive and predictive performance as well as to show practical applications of the proposed models. In particular, a new failure data set collected from a wireless software project is provided in this dissertation for model evaluation. This failure data, in practice, demonstrates the argument of time-delay fault removal assumption since the software faults detected by testers were reported to a centered database system to be removed by the assigned developers after being reviewed. The numerical examples conclude that the consideration of imperfect debugging and delay debugging assumptions can improve the descriptive performance of the models, and in some cases, the predictive performance as well. |
