Relating operational software reliability and workload: results from an experimental study

This paper presents some results from an experimental and analytical study to investigate the failure behavior of operational software for several types of workload. The authors are primarily interested in the so-called highly reliable software where enough evidence exists to indicate a lack of known faults. For purposes of this study, a "gold" version of a well known program was used in the experimental study. Judiciously selected errors were introduced, one at a time, and each mutated program was executed for three types of workload: constant; random; and systematically varying. Analytical analyses of the resulting failures were undertaken and are summarized here. The case when workload is random is of particular interest in this paper. An analytical model for the resulting failure phenomenon based on the Wald equation is found to give excellent results.