Dependability evaluation of software fault-tolerance

The authors present a detailed reliability and safety analysis of the two major software fault-tolerance approaches, recovery blocks (RB) and n-version programming (NVP). The methodology used for the modeling is based on the identification of the possible types of faults introduced during the specification and the implementation, and on the analysis of the behavior following fault activation. The main outcome of the evaluation concerns the derivation of analytical results for identifying the improvement that can result from the use of RB and NVP and for revealing the most critical types of related faults. The study of nested RBs shows that the proposed analysis approach can be applied to such realistic software structures and when an alternate is itself a RB, the results are analogous to the case of the addition of a third alternate. The reliability analysis showed that an improvement has to be expected, but that this improvement would be very low. The study of the discarding of a failed version in NVP shows that this strategy is always worthwhile for safety, whereas, for reliability, it is only worthwhile when independent faults dominate.<<ETX>>