On back-to-back testing

Back-to-back testing involves cross-comparison of all responses obtained from functionally equivalent software components. Whenever a difference is observed it is investigated and, if necessary, a correction is applied. Events associated with back-to-back testing process are defined and examined. The process is first modeled assuming failure independence, and then, assuming failure correlation. It is shown that multiversion testing involving more than four versions, offers rapidly diminishing returns in terms of failure-detection effectiveness, unless additional versions reduce the span of correlated failures. It is also shown that back-to-back testing can remain an efficient way of detecting failures even when the probability of identical and wrong responses from all participating versions, is very close to one.<>