高性能计算中一种改进的数据访问节能技术研究

发布时间：2018-05-08 22:38

本文选题：磁盘功率管理 + 空闲周期　；参考：《计算机科学》2015年02期

【摘要】：针对现有磁盘功率管理方法无法处理高能耗并行应用的短空闲周期、大规模的代码更改和能耗开销较大等问题,提出了一种面向编译器的数据访问调度技术。该技术包括两个阶段:在第一阶段,编译器分析并行应用程序,提取磁盘访问模式,然后生成调度表;在第二阶段,"数据访问调度器"根据调度表进行数据访问。与先前基于软件的策略相比,所提方法不需要改变代码或数据结构。实验评估结果表明,对于数据密集型工作负载,所提方法可以有效提升节能效果,节能率从5.5%上升到11.8%,从而增加了磁盘降速策略对数据密集型高性能计算的可行性。此外,它也将多速磁盘的节能效果从12.7%增加到了27.6%。
[Abstract]:Aiming at the problems that the existing disk power management methods can not deal with the short idle period of high power parallel applications, the large scale code changes and the large energy consumption, a data access scheduling technology for compiler is proposed. The technique consists of two stages: in the first stage, the compiler analyzes parallel applications, extracts disk access patterns, and then generates scheduling tables; in the second stage, "data access schedulers" access data according to scheduling tables. Compared with the previous software-based strategy, the proposed method does not need to change the code or data structure. The experimental results show that the proposed method can effectively improve the energy saving effect for data-intensive workloads, and the energy saving rate increases from 5.5% to 11.8%, thus increasing the feasibility of disk speed reduction strategy for data-intensive high performance calculation. It also increased the energy efficiency of multi-speed disks from 12. 7% to 27. 6%.
【作者单位】：四川民族学院计算机科学系;
【基金】：四川省教育厅自然科学一般项目:量子程序中一种改进的延时估计算法研究(15ZB0332) 四川省教育厅自然科学重点项目:基于粗糙集理论的数据挖掘算法研究(13ZA0136)资助
【分类号】：TP333.35

【参考文献】