基于动态调频的数据中心节能调度策略研究

发布时间：2018-07-02 12:30

  本文选题：数据中心 + 动态调频调压　； 参考：《南京大学》2016年硕士论文

【摘要】：近几年,随着云计算以及高性能计算的普及,数据中心的规模和数量得到了快速的发展,但同时也带来了巨大的电能消耗。据有关资料统计,数据中心所产生的电能消耗已经接近其运行成本的一半,并且在全世界范围内,数据中心所消耗的电能已经占到了全球总用电量的1.3%。因此,数据中心节能问题刻不容缓,已成为目前数据中心研究的一个热点。处理器的动态调压调频技术(Dynamic Voltage and Frequency Scaling, DVFS)是当前服务器上普遍采用的一种节能技术。该技术根据当前服务器上的负载动态的调度任务并且调节频率及电压,使得处理器工作在一个合适的功耗状态下,以节省电能。目前有相当多的工作致力于研究基于DVFS的能量感知的实时任务调度,其目标是希望以最小的能耗代价来完成一批实时任务。在研究过程中,绝大部分工作都以执行时间和截止时间作为任务的服务质量需求(Quality Of Service, QoS)。但是在某些场景中,任务没有固定的执行时间或者任务的执行时间很难被测得。针对这样的任务,也需要根据其特性来做适当的能量感知的调度,以节省电能。本文提出了新的任务模型从另一个更加通用的角度来描述任务的服务质量需求,使其能够描述不能确切知道执行时间的任务。本文提出了最低的有效频率来描述一个任务的服务质量需求。由于系统只知道任务的最低频率要求,而并不知道任务确切的执行时间,因此需要从一个新的角度来衡量任务执行过程中的能耗大小。因此,本文提出了能耗比的概念,用来描述一个任务在不同频率下消耗的能耗与在最大频率下消耗的能耗之比。基于新的任务模型以及能耗比的定义,本文研究了能量感知的任务分配与调度,来最小化服务器完成一批实时在线任务的能耗比,以期最小化总能耗。通过将问题的规约为可变长的背包问题,本文论证了最小化能耗比的任务的分配问题是NP难的,然后基于问题的特点提出了相应的节能任务调度策略。本文所提的任务分配算法将任务分配到当前正在运行的服务器上,尽可能的减少开启的服务器数目,并使得各个任务运行在根据节能的频率以节省能耗。同时,在研究中发现,当有任务结束的时候,处理器上会有一些空闲的计算资源,如果将部分任务进行迁移,能够使得多个处理器都工作在一个更加节能的频率下,从而进一步降低能耗。但是考虑到服务器间任务迁移有较高的代价,而服务器上的核间迁移却几乎不带来额外开销,因此本文中提出了核间任务策略来进一步减少各个服务器上的电能消耗。据我们所有的知识而言,本文是第一个提出通过能耗比来衡量任务在不同频率状态下的耗能情况,并基于能耗比提出任务分配与迁移相结合的能量感知的任务调度。在真实环境下以及模拟环境下的实验中,本文所提的策略在节能方面都要优于其他比较的算法,说明了本文策略在节能方面能够取得较好的效果。
[Abstract]:In recent years, with the popularity of cloud computing and high performance computing, the size and quantity of data centers have developed rapidly, but it also brings great power consumption. According to the statistics, the power consumption generated by the data center is nearly half of its operating cost and is consumed throughout the world. The electric energy has already occupied the 1.3%. of the total electricity consumption in the world, so the energy saving of the data center is urgent and has become a hot spot in the research of the data center. The Dynamic Voltage and Frequency Scaling (DVFS) is a widely used energy-saving technology on the current server. This technology is based on the technology. The load dynamic scheduling tasks on the front server and adjust the frequency and voltage make the processor work in a suitable power state to save power. There is a considerable effort to study real-time task scheduling based on DVFS based energy perception. The goal is to expect to complete a batch of real-time tasks at the minimum cost of energy. In the process of research, most of the tasks are Quality Of Service (QoS) with execution time and deadline (QoS). However, in some scenarios, the task has no fixed execution time or the execution time of the task is difficult to be measured. In this paper, a new task model is proposed to describe the service quality requirements of a task from another more general point of view, so that it can describe the task that is not exactly aware of the execution time. This paper presents the minimum effective frequency to describe the service quality requirement of an appointment. The minimum frequency requirement does not know the exact execution time of the task, so it is necessary to measure the energy consumption in the execution of the task from a new angle. Therefore, the concept of energy consumption ratio is proposed to describe the energy consumption of a task at different frequencies and the ratio of energy consumption at the maximum frequency. The task model and the definition of energy consumption ratio, this paper studies the task allocation and scheduling of energy sensing to minimize the energy consumption ratio of a server to complete a batch of real-time online tasks in order to minimize the total energy consumption. This paper demonstrates that the assignment problem of minimizing the energy consumption ratio is NP difficult by making the specification of the problem to a variable long knapsack problem. And then, based on the characteristics of the problem, the corresponding scheduling strategy of energy saving is proposed. The task allocation algorithm proposed in this paper assigns the task to the currently running server, as much as possible to reduce the number of open servers, and make each task run at the energy saving frequency to save energy. At the end of the task, there will be some idle computing resources on the processor. If a part of the task is migrated, multiple processors can work at a more energy-efficient frequency to further reduce energy consumption. However, considering the high cost of the task migration between the servers, the inter core migration on the server is a few. In this paper, the inter nuclear task strategy is proposed to further reduce the power consumption on each server. According to all of our knowledge, this paper is the first to propose that the energy consumption ratio is proposed to measure the energy consumption of the task at different frequency States, and based on the energy consumption ratio, a combination of task allocation and migration is proposed. In the real environment and the simulation environment, the proposed strategy is better than the other comparison algorithms in the real environment and the simulation environment. It shows that this strategy can achieve better results in energy saving.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP308;TK018

【相似文献】

中国期刊全文数据库 前10条

1 杨建光;路宗雷;;数据中心选址的探讨[J];智能建筑与城市信息;2008年08期

2 张春晖;杜成章;;利用虚拟化技术建设高效节能的数据中心[J];智能建筑;2009年04期

3 丁麒钢;;数据中心节能建设的逻辑与方法[J];智能建筑;2009年09期

4 赖瑞龙;;关于建设节能环保数据中心的探讨[J];海峡科学;2010年10期

5 ;现代数据中心基础设施建设技术年会2011年即将在北京召开[J];电气应用;2011年03期

6 ;现代数据中心基础设施建设技术年会2011年即将在北京召开[J];电气应用;2011年02期

7 赵阳;;数据中心节能降耗实施途径探讨[J];陕西电力;2011年02期

8 李富宇;;高校数据中心建设中的规范遵循[J];中国新技术新产品;2011年09期

9 黄森;潘毅群;;上海某数据中心能效调研分析[J];制冷与空调;2011年02期

10 周伏秋;谷立静;孟辉;;数据中心节能和优化布局研究[J];电力需求侧管理;2011年03期

中国重要会议论文全文数据库 前10条

1 姚，

本文编号：2090122