Dependability evaluation using UltraSAN

Dependability evaluation is an important, but difficult, aspect of the design of fault-tolerant computer systems. UltraSAN, a stochastic activity network-based package for model-based performance, dependability, and performability evaluation of such systems, is presented. The package supports solution by both simulation-based and analytic methods. Of particular importance in the evaluation of dependable systems by analytic means is the control of an often explosive state space growth. UltraSAN has the potential to do this, using recently developed reduced base model construction techniques which exploit symmetries inherent in the SAN structure during construction of the underlying stochastic process representation. Four analytic solvers are available: SOR and LU decomposition for steady-state variables, and uniformization for transient instant-of-time and interval-of-time variables. For models with characteristics that preclude analytic solution, transient and steady-state simulation can be used. The use of UltraSAN and reduced base model construction is illustrated by considering the evaluation of a fault-tolerant distributed system. Model generation via an X Windows based graphical interface, stochastic process construction, and model solution, using numerical methods, are illustrated.