Application-dependent performability evaluation of fault-tolerant multiprocessors

A case study of performance and dependability evaluation of fault tolerant multiprocessors is presented. Two specific architectures are analyzed taking into account system functionality, actual workloads, failures of system components as well as the inter component dependencies. Since the evaluation of such complex systems has to be performed already during the design phase, simulation models are developed and used to provide insight into the system behavior and to uncover weak points and bottlenecks. Object oriented software design and process oriented simulation techniques are used for model construction allowing sophisticated performance and dependability analysis of massively parallel systems.