Operational survivability in gracefully degrading distributed processing systems

The use of experimental methods and statistical analysis techniques to study factors influencing operational survivability in gracefully degrading systems is investigated. Survivability data are generated using a statistically designed experiment in conjunction with a simulation model of network survivability. Thirty two factors having stable regression coefficients are used to identify ten regression models explaining survivability. Influential factors include the distributed system network, the application system, and the distribution policy. Nine factors are found in all models: the number of nodes in the distribution system, distributed system connectivity, module memory requirements, module-to-module interaction frequency, distribution policy, percent of nodes lost, initial assignment results, available processing capacity at the end of the subcase, and the interaction of all application-related variables. Models that are acceptable from both an estimation and prediction viewpoint are developed. Possible commercial and military applications are suggested.