异构并行系统能耗优化分析模型  

Energy Optimization Model for Heterogeneous Parallel System

作　　者：王桂彬[1] 杨学军[1] 唐滔[1] 徐新海[1] 

机构地区：[1]国防科学技术大学计算机学院并行与分布处理国防科技重点实验室,湖南长沙410073

出　　处：《软件学报》2012年第6期1382-1396,共15页Journal of Software

基　　金：国家自然科学基金(60921062;60903059;60903044);国家科技重大专项(2009ZX01036-001-003-001)

摘　　要：随着处理器功耗不断增大,功耗问题逐渐成为高性能计算机系统设计与实现的首要问题.当前,异构系统已成为高性能计算机的发展趋势之一.与传统同构体系结构相比,异构体系结构具有更高的理论峰值性能和能效,但是如何在满足应用性能的条件下充分发掘异构系统的能效优势,仍是一个挑战性问题.通过将应用程序抽象为由串行段和并行段组成的一般程序模型,建立了异构并行系统能耗优化模型.通过分析方法依次给出并行段以及全程序(多程序段)能耗最优时处理器间满足的关系,分别给出了时间约束下能耗最优的处理器频率选择算法.最后,以CPU-GPU异构系统为平台,通过8个典型应用程序验证了方法的有效性.As the processor＇s power consumption continually increases, power has become the most critical problems during the design and implementation of high performance computer （HPC） system. Nowadays, the heterogeneous system has been one important trend for HPC system. Compared with the traditional homogeneous system, the heterogeneous system has a higher theoretical performance and energy efficiency. However, explointing the potential advantage under the performance constraint is still a challenging problem. This work first establishs the energy optimization model for heterogeneous parallel system via abstracting common applications into the general program model which consists of sequential section and parallel section. Through theoretical analysis, the study conclude the relationship among heterogeneous processors for the minimum energy consumption during the single parallel section and full application （including multiple sections） and provide the corresponding algorithms to decide the operating frequencies under the given performance constraint. Finally, the study evaluates of the proposed model with eight typical applications on a CPU-GPU heterogeneous system.

关 键 词：异构系统 低功耗优化 任务调度 动态电压/频率调节 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]