An important factor for optimistic protocol on distributed systems: granularity

Grain size, the amount of computations between communication points, is a quantity to be tuned appropriately depending on the characteristics of the underlying parallel and distributed machines, application problems, and simulation protocols. As target machines for optimistic protocol, the architectural characteristics of a cluster of DEC Alpha workstations are compared to the MasPar MP-2´s in view of parallel logic simulation. We study the effects of varying grain size on several performance metrics when more than one logical processes are assigned to a physical processor on distributed systems. We obtain analytic formulas for the total number of simulation cycles and the total execution time, and find the optimal grain sizes for several benchmark circuits when the number of processors varies. Our experimental results show that the grain size greatly affects the performance of parallel logic simulation on distributed systems, and the effects vary depending on the machine independent factors