基于HPCC和层次分析法的高性能计算系统评价模型

AHPCC:A High Performance Computer System Evaluation Model Based on HPCC and Analytic Hierarchy Process

HPCC(high performance computing challenge)基准是由DARPA的HPCS(high productivity computing system)项目所发布的评价高性能计算系统的测试基准程序,自推出至今,受到工业界和学术界的广泛关注.但是,HPCC仍有不尽如人意之处,主要表现在其测试结果是若干个指标项,需要测试者和决策者根据这些测试指标项进行分析和评估,缺少一个整体的、直观而统一的评价结果.提出一种基于HPCC和层次分析法的高性能计算系统评价模型——AHPCC(a high performance computer system evaluation model based on HPCC),当系统通过运行HPCC得到测试结果后,使用AHPCC模型对这些测试参数按系统应用目标建立层次结构图,并最终计算得到各系统关于特定应用目标的单一分数.以12个已测出HPCC性能参数的系统为例,使用AHPCC模型计算并分析了系统评价结果.实验结果表明,AHPCC模型提供了实际系统的统一而直观的评价指标,其评价结果符合高性能系统的设计和应用特点.

HPCC （high performance computing challenge） benchmark is issued by DARPA HPCS （high productivity computing system） project which is meant to define the program performance boundaries, and it has received broad attention from both industry and academy. But actually, HPCC is not very popular now, which mostly can be attributed to its multiple performance parameters. Because of lacking of a unique and clear evaluation mark, HPCC requests the analyzers and deciders to make a decision manually from the several test results. This paper proposes AHPCC--A high performance computer system evaluation model based on HPCC and analytic hierarchy process. When the candidate systems run the HPCC benchmark set and obtain the evaluation results, AHPCC builds the hierarchy structure graph based on these benchmark results and the application objects, then calculates the related weight vector for the application object, and finally it gives a unique weight for the ultimate application object per candidate system. This paper calculates and analyses the AHPCC results for 12 high performance systems which have been measured by HPCC benchmark set. The experimental results indicate that AHPCC model can efficiently and correctly evaluate the application characteristics of high performance systems.