云数据中心服务器能耗建模及量化计算
2021-05-06周舟袁余俊明李方敏
周舟 袁余俊明 李方敏



摘 要:构建精确的服务器能耗模型有助于资源提供者预测和优化数据中心的能耗. 针对以往数据中心服务器因未考虑“负载的特征”而导致能耗模型精度低的问题,本文提出一种新的能耗建模及量化计算方法,其主要思路如下:根据数据中心服务器所处理任务特征的不同将其分成三类,分别为计算密集型任务、Web事务型任务和I/O密集型任务;针对每一种类型任务,分析其对服务器各部件能耗的影响;利用“主成分分析法”分析各部件参数对能耗的贡献并选择最具代表性的参数,进而结合多元线性回归和非线性回归方法建立能耗模型. 实验结果表明,本文建立的能耗模型预测精度能达到95%以上;与其它模型相比,精度可提高3%左右.
关键词:云计算;数据中心;能耗模型;任务类型;能效优化
中图分类号:TP311 文献标志码:A
Energy Consumption Modeling and Quantitative
Calculation of Servers in Cloud Data Center
ZHOU Zhou1,2,YUAN Yujunming2,LI Fangmin2
(1. College of Information Science and Engineering,Hunan University,Changsha 410082,China;
2. School of Computer Engineering and Applied Mathematics,Changsha University,Changsha 410022,China)
Abstract:Building an accurate energy-consumption model of servers can assist resource providers in predicting and optimizing energy consumption of data center. To address the problem of low accuracy of energy consumption model caused by the failure to consider "load characteristics" of servers in data center, a new energy consumption model and quantitative calculation method are proposed in this paper. The main ideas are summarized as follows: Firstly, we divide the tasks into three classes: CPU intensive task, transactional web task, and I/O intensive task. Then, energy consumption contributions of all components in a server are analyzed. After that, the dominant component parameters of server energy consumption are chosen by using the Principal Component Analysis (PCA), to build a power model through the multiple linear regression method and non-linear regression method. Experimental results show that the prediction accuracy of the proposed energy consumption model can achieve more than 95%. Compared with other energy consumption models, the accuracy can be improved by around 3%.
Key words:cloud computing;data center;energy consumption model;task types;energy efficiency optimization
隨着云计算数据中心的大量新建,数据中心的能耗问题越来越严重. 近期研究显示[1-2]:全球数据中心的总数已超过300万个,耗电量占全球耗电量的1.1%~1.5%. 我国数据中心也发展迅速,总数已达到40万个,年耗电量已超过500亿千瓦,占全国总耗电量的1.5%. 如果以数据中心的PUE(平均电能使用效率)指数来评测,全球先进数据中心的PUE指数为1.2,而我国的PUE指数大于2.2. 与此同时,大量的报告也显示[3-5]:许多高性能数据中心服务器的利用率却远远低于50%,其原因在于数据中心资源未得到“有效”利用. 因此,节能优化算法的提出有助于提高……
