A tool to aid in the design of parallel computers by computing their performance is developed. Currently queueing networks or stochastic Petri nets are the main methods used for computing the performance indices of parallel systems. A method that uses probabilistic automata to model parallel computers will make the tool easy to use, comprehensive and efficient. Probabilistic automata are a modular way of modeling parallel systems. The modeler is relieved from dealing with a global model of a parallel system as the system is divided into modules that are dealt with separately. Problems such as the design of architectures for certain problem domains and the mapping of algorithms to architectures are studied with the aid of such a tool.
