Software reliability modeling
Donald Musa, Anthony Iannino · ACM SIGMETRICS Performance Evaluation Review · 1981
Estimation of software reliability quantities has traditionally been based on stable programs; i.e., programs that are completely integrated and are not undergoing design changes. Also, it is ordinarily assumed that all code is being executed at one time or another and that test or operational results are being completely inspected for failures. This paper describes a method for relaxing the foregoing conditions by adjusting the lengths of the intervals between failures experienced as compensation. The resulting set of failure intervals represents the set that would have occurred for a completely inspected program that was at all times in its final configuration. The failure intervals are then processed as they would be for a stable program. The approach is developed for the execution time theory of software reliability, but the concepts could be applied to many other models as well. Many definitions are given to describe program size variation and associated phenomena. Attention is focused on the special case of sequential integration and pure growth. The adjustment method is described and its benefits in improving the estimation of quantities of interest to the software manager are illustrated.