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

Abstract

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.