面向云计算的能效优先智能路由与管理协同机制研究

发布时间：2019-06-01 20:18

【摘要】：云计算作为一种新的计算服务模式,通过互联网相连接,利用开放的技术和标准把硬件和软件抽象为动态的资源,对外以服务的形式提供给用户。数据中心作为云计算平台的基础依托设施,正在云计算技术的推动下,以前所未有的规模扩张；然而,数据中心的高能耗、低资源利用率、对环境的污染等问题,一直以来极大的制约着数据中心的发展。云计算使存储和计算功能迁移到远程资源,如虚拟服务和存储系统大多承载在数据中心(Data Center, DC)。这种迁移可以带来显著的节能,有效利用当地资源将减少信息、通信和技术(Information and Communication Technology, ICT)40%的温室气体排放量。因此,云服务供应需要谨慎处理传输网络和数据中心这两者带来的能量消耗。本文重新审视云计算背景下的能耗问题,研究云服务配置时的能效智能路由与管理问题：将综合考虑数据中心和传输网络的能耗,重新制定能耗模型。支持云的网络需要以快速、可靠的方式传送大量数据,鉴于基于波分复用(Wavelength Division Multiplexing, WDM)技术的光网络具有高数据速率和低延迟的性能,选择光网络作为传输网是非常合适的。因此本文把注意力放在减少光网络和IT基础设施的整体能量消耗上。首先,介绍了云计算的概念以及数据中心的概念,针对传统数据中心的不足,着重介绍了云计算数据中心具备的特点。详细分析了数据中心网络结构的绿色节能问题。为解决传输网络的能耗问题,引进了IP over WDM网络,并讨论了IP over WDM网络的节能策略。其次,基于云服务配置特有的anycast原理,智能的为用户选择合适的数据中心并进行路由,来提高云服务的能效。针对数据中心和IP over WDM网络的能耗特点,从集中式的角度研究面向云计算的能效优先智能路由与管理协同机制,建立最小能耗MILP模型,目的是通过关闭数据中心和传输网络中不用的资源,来最小化总能耗。由于MILP模型在大规模网络求解方面的复杂性,本文提出了基于进化博弈论的能效优先智能路由光旁路启发式算法,不但考虑尽可能减少虚链路中耗能较大的IP路由器的使用,也考虑关闭数据中心中闲置的服务器,以最小化对某服务传输和处理的总能耗。最后,研究分布式的面向云计算的能效智能路由与管理的联合设计方法,提出能效蚁群算法,通过利用人工蚂蚁实时存放的信息素,及时获知当前不断变化的网络状态,构建分布式的、灵活的、实时的云计算网络能效路由与管理协同机制,联合控制层面与数据层面的网络性能和流量感知,构建高能效的云计算网络智能协同算法。它不使用任何监督,只是让输入流量参照累积的信息素和流量中心性原则流动。然后,输入流量被自动地汇聚在特定链路,且未使用的链路通过数据流逐渐显现。通过关闭未使用的链路实现节能的目的。同时为方便管理,还采用了路由器卡休眠策略,为进一步节能又对路由器端口的分配方式进行研究。通过仿真实验比较分析,本文所提出的两种算法能高效的解决面向云计算的能效路由与管理问题,而且这两种算法在提高网络和IT资源的整体能效、降低能源浪费及提高云服务质量方面体现出了优越性。
[Abstract]:As a new computing service mode, cloud computing is connected through the Internet, and the hardware and software are abstracted into dynamic resources by using open technologies and standards, and are provided to users in the form of services. As the foundation of cloud computing platform, the data center is being expanded with unprecedented scale under the promotion of cloud computing technology; however, the data center's high energy consumption, low resource utilization rate, environmental pollution and other problems have been greatly restricted to the development of the data center. Cloud computing has migrated storage and computing capabilities to remote resources, such as virtual services and storage systems, which are mostly hosted in the data center (Data Center, DC). This migration can lead to significant energy savings and the effective utilization of local resources to reduce greenhouse gas emissions by information, communications and technology (ICT)40 per cent. Therefore, cloud service provisioning requires careful handling of the energy consumption of both the transport network and the data center. This paper reexamines the energy consumption in the background of cloud computing, and studies the energy efficiency intelligent routing and management in the cloud service configuration: the energy consumption of the data center and the transmission network will be taken into account, and the energy consumption model can be reformulated. The network supporting the cloud needs to transmit a large amount of data in a fast and reliable manner, and in view of the high data rate and low latency performance of the optical network based on the wavelength division multiplexing (WDM) technology, it is very appropriate to select the optical network as the transmission network. So this article focuses on reducing the overall energy consumption of the optical network and the IT infrastructure. First of all, the concept of cloud computing and the concept of data center are introduced. In view of the shortage of the traditional data center, the characteristics of cloud computing data center are introduced. The green energy-saving problem of the data center network structure is analyzed in detail. In order to solve the energy consumption problem of the transmission network, the IP over WDM network is introduced, and the energy-saving strategy of the IP over WDM network is discussed. Secondly, based on the ancyast principle specific to the cloud service configuration, the intelligent user selects a suitable data center for the user and routes them to improve the energy efficiency of the cloud service. Aiming at the energy consumption characteristics of the data center and IP over WDM network, the energy efficiency priority intelligent routing and management cooperation mechanism for cloud computing is researched from a centralized point of view, and the minimum energy consumption MILP model is established, and the purpose of the invention is to minimize the total energy consumption by closing the unused resources in the data center and the transmission network. Because of the complexity of the MILP model in large-scale network solution, this paper presents a heuristic algorithm of energy efficiency priority intelligent routing based on evolutionary game theory, which not only considers the use of the IP router with great energy consumption in the virtual link, Also consider shutting down idle servers in the data center to minimize total energy consumption for a service transmission and processing. and finally, the joint design method of the distributed cloud computing-oriented energy efficiency intelligent routing and management is researched, an energy efficiency ant colony algorithm is proposed, The real-time cloud computing network energy efficiency routing and management cooperation mechanism, the joint control layer and the data plane network performance and the flow sensing, and the intelligent cooperation algorithm of the cloud computing network with high energy efficiency is constructed. It does not use any monitoring, just allowing incoming traffic to flow with reference to the accumulated pheromone and flow-center principles. The input traffic is then automatically converged on a particular link and the unused link is gradually visualized through the data stream. The purpose of energy conservation is achieved by closing unused links. At the same time, it is convenient to manage, and also adopts the router card dormancy strategy to study the way of further energy saving and the distribution of the router port. Through the comparative analysis of the simulation experiments, the two algorithms proposed in this paper can solve the energy efficiency routing and management problems facing the cloud computing efficiently, and the two algorithms have the advantages in improving the overall energy efficiency of the network and the IT resource, reducing the energy waste and improving the quality of the cloud.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【相似文献】