A Study Of Amdahl's Law Performance Model On Asymmetric Multicore System

Amdahl's law has been one of the most important parallel computing models in the last forty years. It provides theory for system analysis and design. Besides, it also reveals the performance limitation of application.Parallel theory always bases symmetric multicore architecture whose cores have equal computing performance. But Amdahl's law pointed out the performance will be limited by the serial fraction. To break this limit, a new research direction which called asymmetric multicore system became in computer architecture and processor field. However, there is no research existed to show how to reconstruct the asymmetric architecture.We refine the resource allocation of cores, especially the specific resources on the asymmetric multicore processor, and give a comprehensive model consisting of influential factor on two architectures and their final speedup. Proved by the analysis of model, applications running on the asymmetric processor have higher speedup than running on the symmetric processor in most of the cases. The new model also describes that in order to maximize the overall performance, the performance of big core in asymmetric processor should be designed equal to or better than each core in symmetric architecture.The experiment platform is multi2sim simulation framework. After changing the source code, the simulation framework can support our performance model. The experiment result proves the theory of our study.