基于共享Cache划分的电力芯片能耗优化技术  

Shared Cache partition-based optimization technology for the power chip energy consumption

作　　者：姚浩 黄开天 余宏洲 王轲 YAO Hao;HUANG Kaitian;YU Hongzhou;WANG Ke(Digital Grid Research Intitute,CSG,Guanzhou 510670,China;Electric Power Research Institute,CSG,Guangzhou 510663,China;College of Information Science&Electronic Engineering,Zhejiang University,Hangzhou 310000,China;College of Electrical Engineering,Zhejiang University,Hangzhou 310000,China)

机构地区：[1]南方电网数字电网研究院有限公司,广东广州510670 [2]南方电网科学研究院有限责任公司,广东广州510663 [3]浙江大学信息与电子学院,浙江杭州310000 [4]浙江大学电气工程学院,浙江杭州310000

出　　处：《电力科学与技术学报》2021年第5期28-34,共7页Journal of Electric Power Science And Technology

基　　金：国家重点研发计划(2018YFB0904900,2018YFB0904902)。

摘　　要：提高电力终端芯片工作效率的同时降低其能耗,是优化智能电网系统的研究方向之一。首先针对MPSoC中高速缓存数据的高效管理问题,开展多处理器共享高速缓存划分(CP)技术研究,利用曲线拟合技术对高速缓存建模,通过数学方法求解CP问题;然后基于得到的缺失率曲线,根据共享高速缓存的缺失率与子系统能耗之间的数学关系,得出子系统能耗的数学表达;最后结合处理器能耗模型,综合全局求出最优的CP方案。实验验证表明使用求得的CP方法,处理器子系统能耗是进行优化前的子系统能耗的27.9%。Improving the working efficiency of power terminal chips while reducing their energy consumption is one of the research direction for optimizing smart grid systems.Aiming at the efficient management of cache data in MPSoC,the multi-processor shared cache partitioning technology is studied.The curve fitting technology is utilized to model the cache,and mathematical methods is incorporated to solve the CP problem.The mathematical expression of the energy consumption in subsystem can be obtained according to the mathematical relationship between the obtained missing rate curve of the shared cache and the energy consumption of the subsystem.Combined with the energy consumption model of the processor,the comprehensive optimal CP solution is generated.Experimental verification shows that the processor subsystem energy consumption can be reduced to 27.9%of the subsystem before optimization using this CP method.

关 键 词：共享高速缓存划分技术 缺失率 曲线拟合 子系统能耗 

分 类 号：TM763[电气工程—电力系统及自动化]