(A) white box reliability prediction technique for supporting traceability

Abstract

White box reliability analysis techniques provide useful information to make more precise decisions related to software quality through the identification of critical parts and quantitative estimation of the probability that the given software will not cause any system failure. Although it is cost effective to conduct a reliability analysis in earlier development phases, there are difficulties in predicting the software reliability and tracing potential unreliable parts in early phases due to the unavailability of executable and testable implementation. This paper suggests a white box reliability prediction technique that supports traceability in the design phase. The proposed technique divides the reliability analysis process into six parts and utilizes design artifacts such as UML models. In particular, the technique predicts the system reliability by estimating the method level failure intensity through the busy periods and complexity weight values, and by calculating the probability of an activity being transferred to a complete state using a Markov Chain. This technique supports traceability according to the level of analysis, and hence it is possible to quantitatively identify unreliable parts of the software not only at the component level but also at the user’s system usage level or the system’s activity level. Furthermore, it is possible to simulate how the system reliability will change when a system operational profile is changed, because this technique includes the operational profiles explicitly in the UML models. An experimental study was conducted on safety critical embedded software to evaluate the proposed technique’s applicability, accuracy of prediction, and analysis capability. The results of the study show that the technique can predict software reliability more accurately than the existing model and it effectively identifies sources of unreliability and simulates operational profile changes.