混合供电数据中心能耗优化研究进展

Research Advance on Energy Consumption Optimization of Hyper-Powered Data Center

为支持无处不在的IT服务,数据中心像公路和电网一样已经成为了城市必备的基础设施,兆瓦级的数据中心在吞噬着大量能源的同时也给环境带来沉重压力,因此数据中心能耗优化研究备受关注.因为可再生能源的环境和经济成本低廉,人们更倾向于采用可再生能源与化石燃料混合的方式为数据中心供电.提高可再生能源利用率,降低化石燃料的使用量,是数据中心能耗优化的新目标.然而,可再生能源通常具有间歇性、不稳定性和动态变化性,如何在数据中心中充分利用可再生能源一直是一个难题.该文综述了近年来可再生能源混合供电的数据中心能耗优化的主要研究成果.该文首先总结能耗优化的基本思路,指出从资源层、计算层和服务层均可以实现提高可再生能源利用率和降低能源成本的优化目标;接着描述可再生能源供电模型和IT设备耗电模型,从功率控制和负载均衡两个角度对现有工作加以梳理,逐一分析现有优化技术,并针对每一项优化技术提出进一步的研究思路.最后该文根据现有研究成果,探求采用功率控制、负载均衡的软硬件方法能够实现的最佳优化效果,提出能耗优化的约束问题,包括研究思路、仍然存在的问题和可能的解决办法.

To support ubiquitous IT services, data centers have become an essential infrastructure like roads and power grids for cities. The megawatt-level data centers have devoured a lot of energy while put much pressure on the environment. Researches about energy consumption optimization of data center are concerned by people. Because of the lower environmental and economic costs of renewable energy. Renewable energy is generally defined as energy that is collected from resources which are naturally replenished on a human timescale, such as sunlight, wind, rain, tides, waves, and geothermal heat. People are more inclined to use renewable energy and fossil fuel mix for powering data centers. The new goal of energy consumption optimization is not simple reduce the consumed energy, but to improve renewable energy utilization and reduce fossil energy consumption. However, renewable energy is intermittent, unstable and dynamic changes, hence, how to fully utilize renewable energy in a green data center remains to be a challenge.In this paper, the state - of-the - art of energy consumption optimization in the renewable energy hyper-powered data center are reviewed. The paper first summarizes the basic ideas of the optimization, concludes the approaches to improve the renewable energy utilization and reduce the energy cost in the resource layer, computing layer and service layer of a data center, and explains the basic approaches of matching energy generation and consumption; Secondly, this paper describes the power generation model of renewable energy and power consumption model of IT equipment, in the former the solar energy and wind energy are highlighted, and in the latter the energy consumption model to the tasks, requests and virtual mechanisms are focused; Thirdly, the related works are explained from two aspects： power restriction and load balancing, meanwhile proposes the further research prospects for each optimization technique. The power restriction approaches mainly based on “Dynamic Voltage and Frequency Scaling” or “On-off Algorithm”; while the load balancing is a big topic, it includes temporal load balance （workload scheduling） which balances the task scales among servers in one data center in real-time to adapt the renewable energy, or spatial load balance （geographical load scheduling） which balances the task scales among the geographical difference data centers to adapt their renewable energy locally and respectively. The load in both temporal and spatial load balance could be tasks or virtual machines; Finally, according to the existing research results, the best optimization effects that achieved by the software and hardware optimization approaches are argued. As the goal of maximize renewable energy utilization. The issue is abstracted as “power consumption-time curve” and “power generation-time curve” adaptation problem, which could be solved by transforming the “power generation-time curve” to match the “power generation-time curve”. It would definitely confirmed the “power generation-time curve” if the “power consumption-time curve” could be freely transformed, however it is an ideal situation. There must be context related constraints on energy consumption optimization, such as energy conservation, SLA invariance, response time tradeoff, load migration costs. How to identify this constraint is a difficult problem to be solved. So that the problem of the optimization constraints, including research ideas, problems and possible solutions, are given to facilitate other researchers.