An Experimental Evaluation of Methodological Diversity in Multiversion Software Reliability

N-version programming has long been mooted as a method of improving software reliability. Earlier studies, which generated apparently discouraging results, ooered pessimistic prognostications for this general strategy. However, further study of the problem has both reened the evaluation procedure and revealed new opportunities for improvement. In particular, exploitation of methodological diversity has proved to be productive. In this paper, we continue such studies by introducing a methodological extreme, an in-ductive computing technology, neural computing, to extend an earlier study of Modula2 and Prolog versions of the Launch Interceptor problem. In addition, we examine several diversity measures which are presented as more reliable indicators of potential reliability enhancement than the earlier measures of independence of failure and, subsequently, of probability of two-version joint failure. We also examine several strategies for diversity exploitation in multiversion systems. With the caveat that neural computing technology is far from well-understood, the conclusion is that it does ooer a methodologically diverse approach, but perhaps no more so than the diversity potential of imperative versus declarative languages. But as an inductive computing technology, within which versions are both cheap to produce and speciication independent, neural computing opens several new possibilities for enhancing N-version system reliability. Finally, we conclude that the apparent mismatch between pessimistic predictions and optimistic results in previous multiversion studies is founded on a failure to properly appreciate both the full complexity of the general strategy and the full scope of the options available.