东北大学学报:自然科学版 ›› 2016, Vol. 37 ›› Issue (8): 1089-1094.DOI: 10.12068/j.issn.1005-3026.2016.08.006

• 信息与控制 • 上一篇    下一篇

基于SVM的CPU-GPU异构系统任务分配模型

王彦华, 乔建忠, 林树宽, 赵廷磊   

  1. (东北大学 计算机科学与工程学院, 辽宁 沈阳110819)
  • 收稿日期:2015-05-20 修回日期:2015-05-20 出版日期:2016-08-15 发布日期:2016-08-12
  • 通讯作者: 王彦华
  • 作者简介:王彦华(1982-),女,河北保定人,东北大学博士研究生; 乔建忠(1964-),男,辽宁兴城人,东北大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(61272177); 辽宁省软件系统开发与应用重点实验室项目.

A Task Allocation Model for CPU-GPU Heterogeneous System Based on SVMs

WANG Yan-hua, QIAO Jian-zhong, LIN Shu-kuan, ZHAO Ting-lei   

  1. School of Computer Science & Engineering, Northeastern University, Shenyang 110819, China.
  • Received:2015-05-20 Revised:2015-05-20 Online:2016-08-15 Published:2016-08-12
  • Contact: WANG Yan-hua
  • About author:-
  • Supported by:
    -

摘要: 为了改善异构系统的性能和效率,提出并实现了一个两阶段的任务分配模型.该模型对预分配给CPU和GPU的任务集进行多轮调整,以此最大程度地缩短程序的执行时间.首先,使用支持向量机进行任务预处理,支持向量机将任务分成CPU型和GPU型;然后,根据预处理结果以及处理器的特征和状态,并在对分配集合进行多轮调整后实施实际的任务分配.本模型在具体的异构系统中实现,使用多种基准程序进行检测.实验结果表明,对比其他任务分配算法,本文算法能够使性能获得平均43.54%的提升.

关键词: 图形处理单元, 支持向量机, 异构系统, 机器学习, 任务预处理, 任务分配

Abstract: To improve the performance and efficiency of heterogeneous system, a two-stage task allocation model is proposed and implemented, by which the workload allocated to CPU and GPU is adjusted several times to decrease the execution time to the maximum extent. Firstly, the support vector machine (SVM) is used to classify a task into CPU and GPU in pre-treating. Then,after adjusting the allocation sets several times, the model carries out task allocation in the light of the characteristic and status of processors and the result produced by the first stage. Moreover, a real heterogeneous system is evaluated through several benchmarks on the proposed model. Experimental results demonstrate that the proposed model can achieve an average 43.54% of performance improvement, compared with some of the leading-edge allocation techniques.

Key words: GPU (graphics processing unit), SVM (support vector machine), heterogeneous system, machine learning, task pre-treat, task allocation

中图分类号: