Reliability analysis in N-version programming with dependent failures

Abstract

In fault-tolerant software, to enhance the reliability, one of the major approaches is N-version programming. The modeling studies provided in literature help us to understand under which conditions these approaches increase the software reliability. In this paper, the models given in literature for N-version programming are reviewed. We discuss the deficiency in one of these models trying to improve it to agree with another independent reliability model. Furthermore, we examine how we can combine the dependency concept with the independent reliability model and the subject of how we can extend it for N value. By taking into consideration the dependent failure behavior, a "dependent failure" parameter is introduced to the reliability model in case the events are /spl alpha/-dependent. Here, /spl alpha/ measures the similarity percentage of the input sets on which each pair of versions fail. © 2001 IEEE.