Although N-version programming has been employed in some mission-critical applications, the reliability and fault correlation issues remain a debatable topic in the research community. In this paper, we perform a comprehensive evaluation on our recent project data on N-version programming and present statistical investigations on coincident failures and correlated faults. Furthermore, we compare our project with NASA 4-University project to identify the “variants” and “invariants” with respect to failure rate, fault density, coincident failures, related faults, and reliability improvement for N-version programming. Our experimental results support fault tolerance as an effective software reliability engineering technique.
Xia Cai, Michael R. Lyu, Mladen A. Vouk