面向计算功耗效率的超线程调度研究.pdfVIP

山东农业大学学报 自然科学版( ),2017,48(1):93-99 VOL.48NO.12017 Journal of ShandongAgriculturalUniversity (Natural ScienceEdition ) doi:10.3969/j.issn.1000-2324.2017.01.019 面向计算功耗效率的超线程调度研究 * 林何磊冯国富 冀续烨 陈 明, , , 上海海洋大学 信息学院, 农业部渔业信息重点实验室, 上海 201306 摘 要: 相对于传统计算侧重关注计算性能及并行可扩展性而忽视功耗因素，本文分析Hyper threading服务器的功耗 特性，基于超线程计算单元拓扑感知，引入面向能量效率的超线程调度策略，利用线程亲和性改变线程计算单元分 配顺序以提高系统计算与能量效率，克服操作系统默认线程调度策略单一、呆板的缺陷。实验结果表明，所提方案 可在应用层、运行时支持层和操作系统层三层中灵活配置，以影响不同范围内的计算应用，在提高能量效率的同时， 从计算单元利用角度提高计算资源利用率。 关键词: 超线程 能量效率 计算单元调度 计算单元拓扑感知; ; ; 中图法分类号:TP332 文献标识码:A 文章编号:1000-2324(2017)01-0093-07 Study on Hyper Threading Scheduling towards Power Efficiency of Computing * LIN He-lei,FENGGuo-fu ,JIXu-ye,CHEN Ming College of Information Technology, Key Laboratory of Fisheries Information Ministry of Agriculture/Shanghai Ocean University, Shanghai201306,China Abstract:As traditional computing focuses on performance and parallel scalability while ignoring the power consumption factor, based on Computing unit topology awareness of Hyper Threading, this paper analyzes the power consumption of Hyper threading server, and introduces a Hyper Threading Scheduling Strategy for energy efficiency. By setting threads processor affinity to change the allocation order of computing units, the proposed method improves the system computing and energy efficiency, and overcomes the shortage of traditional OS whose schedule strategy is too monotonous for Hyper Threading. The experimental result shows that the proposed scheme can be deployed flexibly in three layers: application layer, runtime library layer and operating system kernel layer to affect computing applications in different sco