摘要
数据中心的一个重要任务是功率控制,功率封顶是数据中心对服务器设置功率消耗上限的技术。关注的是数据中心节点机一级的动态功率控制机制。基于系统级功率模型构建了进程级功率模型,并将两者整合构成"软功率计",用于监控系统功率与进程功率;为实现功率封顶,软功率计被集成到闭环控制系统中,设计了功率控制的算法,该算法在控制系统功率不超预算的情况下,系统以较好的性能运行。实验结果表明所提控制机制能有效地控制系统的实时功率,且性能下降较少,同时也可以改善能效,可应用于功率感知的DBMS服务器中。
Power control is a critical issue in data center and power capping is the technique to keep the system within a fixed power constraint. This paper focused on the dynamic power control scheme in a data center node machine. We constructed a process-level power model based on our previous system-level power model,and integrated these two models into a gadget called soft power-meter to control system power and process power. To achieve power capping, the soft power-meter is integrated into a closed-loop control system, and a power control algorithm is devised, which keeps the system within the fixed power budget with good performance. The experiment results demonstrate that the proposed power capping scheme can effectively control the system's power with small performance degradation, and improve the energy-efficiency. It can be applied to a power-aware DBMS server.
出处
《计算机科学》
CSCD
北大核心
2015年第B11期490-496,共7页
Computer Science
基金
浙江省基金项目(LY14F020017
LY13F020026)
国家自然科学基金面上项目(61070042)资助
关键词
功率封顶
功率建模
软功率计
功率感知数据库系统
Power capping, Power modeling, Soft power-meter, Power-aware database systems