智能电网环境下基于价格的数据中心电力成本优化
发布时间：2018-10-18 08:21
【摘要】：近些年来,随着云计算的普及,网络需求得以快速发展,由数据中心或者数据服务器产生的电力花费也持续增长,并且呈现一种急剧增长的趋势。一项调查报告指出,全球范围内服务器的电力花费可能已经超过了服务器等硬件设备的花费。因此,对降低电力花费进行研究就迫在眉睫了。同时,随着智能电网的发展,越来越多的电网开始施行动态电价机制,特别是分时电价和实时电价。动态电价不仅在时间上具有差异,而且在空间上也具有差异性。当数据中心或者数据服务器配备了电能存储设备,例如电池,就可以在低电价时存储电能,在高电价时释放电能。于是在这种背景下,研究怎样利用电价的特点对电力进行调度分配以减少整个数据中心或服务器的电力成本就显得尤为可行和重要了。 一方面,对数据中心能耗的研究已经取得了很多重要的成果。实际上,它也是一种间接降低电力成本的方法。比较成熟的方法有动态电压调整、动态电压与频率调整、动态调整服务器状态、虚拟机技术等等。另一方面,智能电网的发展带来的动态电价机制,使得数据中心可以迁移网络负载到低电价阶段(或地点)执行或者储存低电价阶段(或地点)电能在高电价阶段(或地点)使用,从而直接降低数据中心的电力成本。 本文基于动态电价重点探讨了三个层面的问题(由浅入深,由简单到复杂)：单数据服务器的电力成本优化、服务器集群的电力成本优化以及分布式数据中心的电力成本优化。针对单数据服务器的电力成本优化问题,考虑了网络负载的随机特性和分时电价的时域性差异,提出动态规划的解决方案,克服了将网络负载作为确定性负载进行处理的缺陷；针对服务器集群的电力成本优化问题,考虑结合通过直接减少电力成本(利用分时电价的时域性差异)和通过节能间接减少电力成本两个方面,提出建立马尔科夫决策过程的解决方案,将以往研究的两种方法进行了针对性的结合；针对分布式数据中心的电力成本优化问题,考虑到了电价的空间和时间差异和网络负载的调度以及网络带宽的限制,同时保障服务质量的需求,提出了优化网络负载调度的解决方案。需要说明的一点：我们所指的电力成本仅是指服务器的电力花费。
[Abstract]:In recent years, with the popularity of cloud computing, network demand has developed rapidly, and the power cost generated by data centers or data servers has continued to increase, and has shown a trend of rapid growth. Worldwide, servers may be spending more on power than on hardware such as servers, according to a survey. Therefore, it is urgent to study the power cost reduction. At the same time, with the development of smart grid, more and more power grids begin to implement dynamic pricing mechanism, especially time-sharing price and real-time price. Dynamic electricity price is not only different in time, but also different in space. When a data center or a data server is equipped with a power storage device, such as a battery, it can store electricity at a low price and release it at a high price. Under this background, it is particularly feasible and important to study how to make use of the characteristics of electricity price to dispatch and distribute electricity to reduce the power cost of the whole data center or server. On the one hand, the research on data center energy consumption has made a lot of important achievements. In fact, it is also an indirect way to reduce the cost of electricity. More mature methods include dynamic voltage adjustment, dynamic voltage and frequency adjustment, dynamic adjustment of server state, virtual machine technology and so on. On the other hand, the dynamic pricing mechanism brought about by the development of smart grid, The data center can migrate the network load to the low price stage (or location) to execute or store the low electricity price stage (or location) for use in the high price stage (or location), thus directly reducing the power cost of the data center. Based on dynamic electricity pricing, this paper focuses on three aspects (from simple to complex): single data server power cost optimization, server cluster power cost optimization and distributed data center power cost optimization. Considering the stochastic characteristics of network load and time-domain difference of time-sharing price, a dynamic programming solution is proposed to solve the power cost optimization problem of single data server, which overcomes the defect of treating network load as deterministic load. Aiming at the problem of power cost optimization in server cluster, we consider combining two aspects: direct reduction of power cost (using time-domain difference of time-sharing price) and indirect reduction of power cost through energy saving. This paper proposes a solution to establish Markov decision process, combining the two methods studied in the past, aiming at the power cost optimization problem of distributed data center. Considering the space and time difference of electricity price, the scheduling of network load and the limitation of network bandwidth, and at the same time guaranteeing the requirement of quality of service, the solution of optimizing network load scheduling is put forward. One point to note: we refer to the cost of electricity only for the server.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP308
本文编号：2278545
[Abstract]:In recent years, with the popularity of cloud computing, network demand has developed rapidly, and the power cost generated by data centers or data servers has continued to increase, and has shown a trend of rapid growth. Worldwide, servers may be spending more on power than on hardware such as servers, according to a survey. Therefore, it is urgent to study the power cost reduction. At the same time, with the development of smart grid, more and more power grids begin to implement dynamic pricing mechanism, especially time-sharing price and real-time price. Dynamic electricity price is not only different in time, but also different in space. When a data center or a data server is equipped with a power storage device, such as a battery, it can store electricity at a low price and release it at a high price. Under this background, it is particularly feasible and important to study how to make use of the characteristics of electricity price to dispatch and distribute electricity to reduce the power cost of the whole data center or server. On the one hand, the research on data center energy consumption has made a lot of important achievements. In fact, it is also an indirect way to reduce the cost of electricity. More mature methods include dynamic voltage adjustment, dynamic voltage and frequency adjustment, dynamic adjustment of server state, virtual machine technology and so on. On the other hand, the dynamic pricing mechanism brought about by the development of smart grid, The data center can migrate the network load to the low price stage (or location) to execute or store the low electricity price stage (or location) for use in the high price stage (or location), thus directly reducing the power cost of the data center. Based on dynamic electricity pricing, this paper focuses on three aspects (from simple to complex): single data server power cost optimization, server cluster power cost optimization and distributed data center power cost optimization. Considering the stochastic characteristics of network load and time-domain difference of time-sharing price, a dynamic programming solution is proposed to solve the power cost optimization problem of single data server, which overcomes the defect of treating network load as deterministic load. Aiming at the problem of power cost optimization in server cluster, we consider combining two aspects: direct reduction of power cost (using time-domain difference of time-sharing price) and indirect reduction of power cost through energy saving. This paper proposes a solution to establish Markov decision process, combining the two methods studied in the past, aiming at the power cost optimization problem of distributed data center. Considering the space and time difference of electricity price, the scheduling of network load and the limitation of network bandwidth, and at the same time guaranteeing the requirement of quality of service, the solution of optimizing network load scheduling is put forward. One point to note: we refer to the cost of electricity only for the server.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP308
【参考文献】
相关期刊论文 前3条
1 张钦;王锡凡;王建学;;尖峰电价决策模型分析[J];电力系统自动化;2008年09期
2 廖家平,方娜,陈红;电力市场条件下分时电价分析[J];东北电力技术;2005年04期
3 崔旭;;智能电网发展现状与节能潜力研究[J];科技经济市场;2012年06期
，本文编号：2278545
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2278545.html
