面向云计算的能效优先智能管理策略与技术研究

发布时间：2019-07-10 12:01

【摘要】：在IT领域,各种“绿色”计算技术也开始进入科研人员的研究范围。近些年云计算作为一种崭新的计算模式受到产业界和学术界的广泛关注。随着用户对云计算需求的不断增长,云数据中心规模日益庞大,其能耗问题也越来越严重,全球数据中心在2005年到2010年增加了约56%,美国数据中心的能耗增加了36%。一个5万个节点的数据中心每年消耗的电能超过1亿千瓦时,能耗占了一个数据中心运维成本中的40%。在国内,通信运营商的数据中心是耗电大户。到2020年,全球主要云计算运营商的能耗将接近2万亿千瓦时。因此,研究云环境下的节能技术有迫切的需求。云计算能耗问题逐渐成为制约云计算发展的重要因素,如何降低网络能耗,同时提高网络的整体能效是本文的研究重点。研究发现,现行网络中20%的链路承载着80%的流量负载,这说明大部分网络的设计都是不合理的。另外,大部分网络都是采用冗余设计来满足突发流量、链路拥塞等网络突发问题,但是突发问题发生的概率小,却因此浪费了大量的资源。且大部分的网络节能算法都是单纯为了节能而设计的,没有考虑休眠链路后网络的性能是否会降低,如何提高网络能效,同时尽量保证网络的性能变化最小是本文的研究目的。本文基于多目标优化理论和进化算法,在控制层面与数据层面分离的背景下,对适用于不同规模的云计算网络的能效路由策略、休眠策略问题进行了研究,在保证QoS的基础上,优先提高网络能效。本文主要通过两方面实现能效网络,一是能效路由策略,二是休眠策略。首先根据最小化网络比特能耗及延迟设计了多目标优化模型,提出了基于离散粒子群的能效优先路由算法(S-PSO-EERA),并通过重路由策略保证了网络的链路最高利用率限制,降低了请求阻塞数。然后,根据最小化网络比特能耗最大化网络休眠链路数设计了多目标优化模型,提出了基于多目标遗传算法和流重定向的多目标能效优先路由算法(NSGAII-FR-EERA)。算法内容包括基于多目标遗传算法的能效路由策略(NSGAII-EERA)和基于流重定向的链路休眠策略(FR-LSA)两部分。最后通过仿真实验针对不同规模网络、不同流量负载变化的情况分析了网络的能效情况,以及网络的性能,并与经典路由算法、休眠算法进行了对比分析。仿真实验表明,本文提出的两种基于进化算法的多目标能效优先路由算法具有可行性、有效性和可扩展性。
[Abstract]:In the field of IT, various "green" computing technologies have also begun to enter the research scope of researchers. In recent years, cloud computing, as a new computing model, has been widely concerned by industry and academia. With the increasing demand of users for cloud computing, the scale of cloud data center is becoming larger and larger, and its energy consumption problem is becoming more and more serious. The global data center increased by about 56 percent from 2005 to 2010, and the energy consumption of data center in the United States increased by 36 percent. A data center with 50,000 nodes consumes more than 100 million kilowatt-hours of electricity a year, accounting for 40 percent of the operation and maintenance costs of a data center. In China, the data center of communication operators is a large power consumer. By 2020, the world's leading cloud computing operators will consume nearly 2 trillion kilowatt-hours. Therefore, there is an urgent need to study energy saving technology in cloud environment. Cloud computing energy consumption has gradually become an important factor restricting the development of cloud computing. How to reduce network energy consumption and improve the overall energy efficiency of the network is the focus of this paper. It is found that 20% of the links in the current network carry 80% of the traffic load, which indicates that the design of most networks is unreasonable. In addition, most networks use redundant design to meet burst traffic, link congestion and other network burst problems, but the probability of burst problems is small, so a lot of resources are wasted. And most of the network energy saving algorithms are designed only for energy saving, without considering whether the performance of the network will be reduced after the dormant link, how to improve the network energy efficiency, and how to ensure the minimum performance change of the network is the purpose of this paper. Based on the theory of multi-objective optimization and evolutionary algorithm, this paper studies the efficient routing strategy and dormancy strategy suitable for different scale cloud computing networks under the background of separation of control level and data level. On the basis of ensuring QoS, priority is given to improving network energy efficiency. This paper mainly realizes energy efficiency network through two aspects, one is efficient routing strategy, the other is dormancy strategy. Firstly, a multi-objective optimization model is designed according to minimizing the bit energy consumption and delay of the network, and an energy efficiency priority routing algorithm based on discrete particle swarm optimization (S-PSO-EERA) is proposed, and the maximum utilization limit of the network link is guaranteed by rerouting strategy, and the number of request blocking is reduced. Then, a multi-objective optimization model is designed according to the minimum network bit energy consumption maximization network dormancy link number, and a multi-objective energy efficiency priority routing algorithm (NSGAII-FR-EERA) based on multi-objective genetic algorithm and flow redirection is proposed. The algorithm includes two parts: efficient routing strategy based on multi-objective genetic algorithm (NSGAII-EERA) and link dormancy strategy based on flow redirection (FR-LSA). Finally, the energy efficiency of the network and the performance of the network are analyzed by simulation experiments for different scale networks and different traffic loads, and compared with the classical routing algorithm and dormancy algorithm. The simulation results show that the two multi-objective energy efficiency priority routing algorithms based on evolutionary algorithm are feasible, effective and extensible.
【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】：TP3

【相似文献】