Summary: | 碩士 === 元智大學 === 資訊工程學系 === 104 === Currently software reliability is generally calculated by collecting data during testing stages and using these data as parameters in various reliability growth models for calculation. However, program structures and behavior are not analyzed in such an approach. Thus, it is difficult for software reliability to get consensus in industry and academia.
This study applies hardware failure physics to software and proposes a white-box approach with 7 levels to analyze software reliability. Our model includes not only experiments but also theoretical aspects to analyze program internal structures. Besides, we also discuss the needed numbers of experiments in order to obtain the failure rates calculated from the theoretical part in our model. The confidence of the experiment results is thus enhanced by the theoretical support and the calculated numbers of experiments in this model.
Three case studies (Partition of Quicksort, Pentinum Division Error, Mars PathFinder Incident) are conducted to demonstrate the effectiveness of our proposed approach Discussions and insights are also given.
|