Execution-driven simulation of multiprocessors

Abstract

This article describes and evaluates an efficient execution-driven technique for the simulation of multiprocessors that includes the simulation of system memory and that is driven by real program work loads. The technique produces correctly interleaved address traces at run-time without disk access overhead or hardware support, allowing accurate simulation of the effects of a variety of architectural alternatives on programs. We have implemented a simulator based on this technique that offers substantial advantages in terms of reduced time and space overheads when compared to instruction-driven or trace-driven simulation techniques, without significant loss of accuracy. The article presents the results of several validation experiments used to quantify the accuracy and efficiency of the simulator for sequential, distributed, and shared-memory multiprocessors, and several parallel programs. These experiments show that prediction errors of less than 5 percent as compared to actual execution times, and overheads 6 to 30 times lower than those incurred by cycle-level simulation can be achieved. Predictions of relative performance metrics such as speedup tend to be even more accurate, making this technique especially attractive as an efficient method for comparative investigations of parallel system designs.