面向能耗优化的数据中心资源动态调度模型与方法

发布时间：2018-05-10 19:38

本文选题：数据中心资源 + 动态调度　；参考：《哈尔滨工业大学》2013年硕士论文

【摘要】：随着社会经济、科技的不断发展与进步，IT行业与人们工作、生活的关系越来越密切。IT行业的蓬勃发展使得数据中心的数量越来越多，但是大小林立的数据中心在为人们提供服务的同时，能耗问题也变成了一个日渐沉重的话题。当今时代是一个节能的时代，是我国重要的长期战略任务。数据中心能耗量的大幅度增长，已经成为供电能力的最大威胁。因此，数据中心的节能问题是当前的重点需要解决问题。那么，，数据中心主要的耗电资源设施成为了我们研究的重点。由于数据中心资源类型种类较多并且规模庞大，并且需要满足用户需求的不确定性。在这种情况下如果不对资源进行合理的控制与调度，就会造成电能的浪费。因此，以优化能耗为目标的数据中心资源动态调动问题是本文的主要研究内容。本文首先对数据中心资源调度问题的国内外研究现状进行了分析，总结了已有的成果，提出了值得探索的研究方向，并且简要介绍了目前数据中心资源能耗仿真软件的研究情况。文章从数据中心资源模型的建立入手，将资源按照用途划分为计算资源与物理资源，并且对主要的能耗设施的工作原理进行了详细的分析，建立了各类资源的基本模型，详细描述了数据中心资源对外提供的服务以及资源之间的交互关系。文章根据各类资源的特点与用途，建立了数据中心的能耗模型，提出了数据中心资源动态调度的问题，并对此调度问题进行了定义。针对数据中心资源调度的特点，本文给出了基于CPS的分层调度模型。由于数据中心的资源具有多类型、多属性的特点，文章采用基于近似动态规划的方法进行资源动态调度。文章首先对近似动态规划的基本概念、求解思想与求解过程进行了详细的描述分析。然后根据近似动态规划的建模思想建立了数据中心资源动态调度模型，根据此类问题的特点，提出了求解问题的启发式规则，并且结合问题模型与求解规则提出了此问题基于近似动态规划算法的解决方法，并对算法进行了训练阶段的实验。文章实现了一个数据中心资源能耗仿真软件，将EnergyPlus与Green CloudSimulator进行了集成，并对训练结果的实验结果在此平台上进行了应用仿真实验，验证了算法的正确性与有效性，并且与其他的调度算法做了比较，验证了基于近似动态规划数据中心资源动态调度算法的优越性。
[Abstract]:With the development of society and economy, the continuous development and progress of science and technology, the relationship between IT industry and people's work, the flourishing development of IT industry makes the number of data centers more and more. But while large and small data centers serve people, energy consumption has become a growing issue. Today is an energy-saving era, is an important long-term strategic task of our country. The large increase of energy consumption in data center has become the biggest threat to power supply capacity. Therefore, the data center energy-saving problem is the current key need to solve the problem. So, the main power-consuming facilities in the data center have become the focus of our research. Because of the variety of data center resource types and the large scale, it needs to meet the uncertainty of user needs. In this case, if the resources are not properly controlled and scheduled, it will lead to the waste of electricity. Therefore, the dynamic mobilization of data center resources aiming at optimizing energy consumption is the main content of this paper. Firstly, this paper analyzes the domestic and international research status of data center resource scheduling problem, summarizes the existing results, and puts forward the research direction that is worth exploring. And briefly introduces the current data center energy consumption simulation software research situation. Starting with the establishment of the resource model of the data center, the paper divides the resources into computational resources and physical resources according to their use. The working principle of the main energy consumption facilities is analyzed in detail, and the basic models of various kinds of resources are established. The service provided by data center resources and the interaction between them are described in detail. According to the characteristics and uses of all kinds of resources, the energy consumption model of data center is established, the problem of dynamic scheduling of data center resources is put forward, and the scheduling problem is defined. According to the characteristics of data center resource scheduling, this paper presents a hierarchical scheduling model based on CPS. Due to the multi-type and multi-attribute characteristics of the resources in the data center, this paper adopts a method based on approximate dynamic programming for the dynamic scheduling of resources. In this paper, the basic concept of approximate dynamic programming, the idea of solving and the process of solving are described and analyzed in detail. Then, according to the modeling idea of approximate dynamic programming, the dynamic scheduling model of data center resources is established. According to the characteristics of this kind of problems, a heuristic rule is proposed to solve the problem. A solution based on approximate dynamic programming algorithm is proposed based on the problem model and solution rules, and the algorithm is tested in the training stage. In this paper, a data center resource energy consumption simulation software is implemented, which integrates EnergyPlus and Green CloudSimulator. The simulation results of the training results are applied on this platform to verify the correctness and effectiveness of the algorithm. And compared with other scheduling algorithms, the superiority of resource dynamic scheduling algorithm based on approximate dynamic programming data center is verified.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP308;TP301.6

【相似文献】