基于CC-NUMA系统模拟器的并行程序性能分析

摘要

针对CC-NUMA并行系统的特点，本文描述了模拟器—AMY的设计与实现。该模拟器运行在x86 PC机上的Linux操作系统环境下，采用多项优化技术，能够较精确地统计并行程序的时间开销和CC-NUMA并行系统的各项参数，具有执行速度快、精度高和内存开销小等特点。在AMY模拟器环境下，通过对几个典型的并行测试程序的模拟执行，文章给出了统计的模拟结果，分析了并行测试程序的执行行为和开销，最后得出了在CC-NUMA并行系统中对并行程序进行性能优化的有益的指导原则。%According to the characteristics of the CC-NUMA parallel system,this paper introduces a simulator for the CC-NUMA system AMY that runs on the Linux Operating System in common x86 PCs.AMY adopts many optimization technologies,so it can state the time overhead of parallel programs and statistic parameters of CC-NUMA very fast and accurately with little memory overhead.According to the execution of some parallel benchmarks on AMY,this paper state the simulated results,analyzes the behavior and overhead of parallel benchmarks and concludes the helpful guideline of optimizing parallel programs in the CC-NUMA parallel system.