面向数据中心的服务器能耗模型综述

A Survey of Server Energy Consumption Models in Data Center

摘要:

伴随着云计算技术的快速发展,数据中心的服务器能耗日益激增,带来了严重的经济和环境问题,降低数据中心能耗,对缩减数据中心运营成本、实现全球“双碳”战略目标具有重要意义。因此,不同层面的服务器能耗模型构建和预估成为了近年来研究的热点。据此,从硬件、软件层面系统地总结了服务器能耗模型的相关工作。在硬件层面,对服务器的整体能耗按加法模型、基于系统利用率模型和其他模型分类；同时,还总结了服务器部件粒度的能耗模型,涵盖CPU、内存、磁盘和网络接口。在软件层面,按机器学习的类别将服务器能耗模型归纳为监督学习、非监督学习、强化学习。此外,还比较了不同能耗模型的优缺点、适用场景,展望了能耗模型的未来研究方向。

Abstract:

With the rapid development of cloud computing, the increasing demand for server energy consumption in data center leads to crucial economic and environmental issues. Reducing the data center energy consumption is of great significance to cut down the operating cost of data center and realize the global "double-carbon" strategic goal. Therefore, an increasing amount of research on power consumption models and prediction at different levels in cloud servers. This paper conducted a systematic study about existing work in power consumption models from two levels, hardware and software. At the hardware level, the overall energy consumption models of the cloud server is classified according to the additive server power models, system utilization based server power models and other server power models, the energy consumption models of the server components are also presented, including the CPU, memory, disk and network interface. At the software level, the server energy consumption models are summarized according to the category of machine learning, such as supervised learning, unsupervised learning and reinforcement learning. By comparing existing approaches and solutions, we analyzed their advantages, limitations, and suitable scenarios. In addition, we also pointed out several possible research directions.