This Small Business Innovation Research Phase I project will investigate practical techniques for software reliability amplification. Reliability-amplification occurs when some aspect of the software development process is shown to complement reliability testing, in the sense that predictions of the software failure rate are demonstrated to be pessimistic. The main objective of this effort is to determine whether it is feasible to provide a framework for "reliability amplification" as a new quantitative approach to software validation that establishes statistical confidence in software quality. Since reliability amplification rests on a failure-rate bound prediction, and there is no plausible statistical model of software failure, theoretical work is obviously required in this area. Two approaches are promising: (1) Study of the software computation as opposed to the input/output behavior. The computation contains full details of the internal software state, and may be an appropriate home for a failure rate more plausible than one assumed for the input domain. (2) Input subdomains homogeneous for failure. If it is not plausible that a constant, failure rate exists across the input space, perhaps there is a partition of that space in whose subdomains constant failure rates are more believable. The results of this effort will be a report detailing the feasibility of building a new software reliability framework that is based on defect-removal methods as opposed to purely testing methods. Applications of this research include defense, medical, aerospace, nuclear telecommunications, and any other application requiring extremely high software reliability assessments.
