The basic research component of this project is concerned with transient faults and unexpected environmental changes that occur in distributed systems and networks. Transient faults include spontaneous data corruption and any temporary deviation from software specification. The project explores how effects of transient faults can be contained within desired boundaries of time or application-specific objectives. Using methods of stabilizing algorithms, which automatically recover from transient state disruptions, issues of containment will be studied. Basic research investigations will yield new algorithms for tolerating transient faults and will demonstrate the inherent resource conflicts between stabilization and containment objectives. Another goal is to alleviate compute-intensive and complexity disadvantages often associated with stabilizing algorithms, using methods of encapsulation and probabilistic approaches. The potential significance of the new stabilizing containment techniques will be demonstrated by case studies: interfaces between software layers that contain effects of faults are simplified; optimistic treatment of faults improves efficiency of recovery; distributed programs can manipulate objects while limiting contamination due to temporary object corruption. The impact of the basic research and its case studies will be to bring stabilization and containment techniques closer to widespread application in new generations of distributed systems. The plan's applied research and educational components emphasize student projects implementing recent distributed computing algorithms.
