摘要
分析了几种可扩展性能评价模型存在的问题,针对实际评价与测试的需要,提出了一种基于等平均负载的数值并行计算可扩展性评价模型.该评价模型对可扩展性能加速比和可扩展性进行了重新定义,给出了使用该模型的进行可扩展加速比和可扩展性测试的方法,结合曲线拟合或并行计算时间模型可以预测并行系统的可扩展性,对NPBBT,SP和矩阵乘法进行了可扩展性预测.
In this paper, the existing problems of the current scalability models are analyzed, and aiming at the requirements of the practical evaluations and tests, a practical scalability metric based on iso-average-computation-load is proposed to provide a quantitative measurement of the scalability. In this metric, the definitions of scalable speedup and scalability are different from the current metrics. By using this metric, a practical method can be obtained to test scalable speedup and scalability, and combined with curve fitting or parallel computing time model, the scalabilities of parallel systems can be predicted.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2005年第6期1073-1078,共6页
Journal of Computer Research and Development
基金
国家自然科学基金项目(40245023)
计算物理国家重点实验室基金项目(51479040103KG0201)
关键词
数值并行计算
可扩展性
加速比
性能评价模型
numerical parallel computation
scalability
speedup
performance evaluation metric
