The discipline of discrete event simulation may be utilized to model many physical systems such as digital hardware, queueing networks, telephone networks, simulated warfare, and banking transactions. Where the entities of a physical system execute independently and interact asynchronously, an asynchronous distributed event-driven simulation algorithm may enable the simulation of the system to execute on a parallel processor. This has the potential to significantly reduce the total simulation time. YADDES is the first algorithm that is characterized by: (1) acceptable performance, (2) freedom from deadlock, and (3) probably correct, for circuits where the interactions between the entities constitute a cyclic dependence. Given their complex, asynchronous nature, an important issue associated with all asynchronous distributed algorithms is their correctness, i.e., the generation of accurate output for given input stimulus, under all possible conditions. This paper presents a mathematical proof of correctness and reports the performance of YADDES on the Armstrong parallel processor at Brown University.
Read full abstract