Exhaustively testing software is generally infeasible, testing software to high levels of reliability is intractable, and software assurance models are often viewed with suspicion. Yet software production forges ahead, resulting in the most complex artifacts in human history: modern software systems. As far as assessing the ``goodness'' of software goes, we are facing greater challenges today than we faced a decade ago. Unfortunately, we cling to the only mildly effective techniques of the 80's. This talk asks the reader to take a critical look at the positive impact that software fault-injection can make toward measuring software risk. Software fault-injection is an emerging technology that can be used to observe how software systems behave under experimentally-controlled, anomalous circumstances. Software fault-injection acts as a crystal ball, predicting how badly software might behave should things go awry (both internally and externally) during execution. Such predictions provide clues as to how robust a piece of code is, where in the code failure tolerance is deficient, and most importantly, what level of risk is incurred by relying on a particular software system. In short, fault-injection is an efficient means for predicting the future and counterfactualizing about specific possible events. Depending on time, 3 case studies will be described during the talk that have successfully employed fault-injection: (1) Bay Area Rapid Transit, (2) Magneto Stereo Taxis System, and (3) TCASII. Two ongoing efforts for which preliminary results may be ready by the time of the talk involve WAAS and the Halden Nuclear Reactor Project in Norway.

Biography

Jeffrey Voas is a research scientist at Reliable Software Technologies (RST) Corporation, and is currently the principal investigator on research initiatives for DARPA, National Institute of Standards and Technology, and National Science Foundation. He has published over 75 journal and conference papers in the areas of software testability, software reliability, debugging, safety, fault-tolerance, design, and computer security. Before co-founding RST, Voas completed a two-year post-doctoral fellowship sponsored by the National Research Council at the National Aeronautics and Space Administration's Langley Research Center. Voas has served as a program committee member for numerous conferences, and will serve as Conference Chair for COMPASS'97. Voas has coauthored a text entitled Software Assessment: Reliability, Safety, Testability (John Wiley & Sons, 1995 ISBN 0-471-01009-X). Voas is currently writing a successor book to Software Assessment devoted solely to the topic of software fault-injection. In 1994, the Journal of Systems and Software ranked Voas 6th among the 15 top scholars in Systems and Software Engineering. Voas received his Ph.D. in computer science from the College of William & Mary in 1990.

CREST home page	News	Research	Seminar Series	Publications	Feedback