基于云计算环境的资源提供优化方法研究

发布时间：2018-07-09 18:20

本文选题：云计算 + 节能　；参考：《武汉理工大学》2013年博士论文

【摘要】：作为一种新兴的信息处理模式,云计算(Cloud Computing)技术已经成为信息领域备受关注的研究热点。云计算以虚拟化(Virtualization)作为支撑技术,以按需方式向Internet用户提供动态可扩展的服务。然而,由于云计算环境规模大,资源管理与分配动态可伸缩的特点,导致云数据中心的能耗问题及其资源提供效率成为影响云计算性能的关键因素。本文以新的计算基础设施——云计算技术为背景,研究如何优化云计算数据中心的能耗及其资源的优化配置问题。到目前为止,云计算的能耗问题及其资源提供依然存在很多亟待解决的问题。本文重点从节能机制、负载均衡和市场经济模型等方面研究云计算环境中的高效资源提供优化方法,主要的研究工作包括以下几点： 1)系统研究了云计算环境中的节能机制及其资源提供优化方法。首先,从云计算的基本概念入手,介绍了云计算的特点、服务类型及层次；其次,重点研究了云计算中的节能优化策略,分析比较了策略的应用环境及优缺点；然后,进一步研究了云计算中的资源提供技术,并对该领域目前的优化策略进行了分类比较；最后,对云实验环境CloudSim进行介绍并对其资源提供机制进行实验分析。 2)提出了基于能量与SLA均衡的虚拟机资源提供策略。针对云计算环境中应用需求的动态变化特性,提出了基于强局部加权回归的虚拟机自适应部署算法RLWR, RLWR可以根据应用负载所体现的资源占用历史信息动态决策主机的超载时机。检测出超载主机后,提出了迁移周期最优的虚拟机迁移选择算法MPM和迁移量最小算法MNM进行迁移虚拟机的选择,然后提出以基于功耗的降序最佳适应启发式算法PBFDH对迁移虚拟机进行再次优化部署。该自适应部署策略比较静态阈值算法STH、MPA和DVFS,不仅可以动态地将虚拟机部署到更少物理主机上,从而关闭闲置主机,提高了能效,而且通过主机资源的负载预测实现了高可靠的QoS服务交付,避免了用户与资源提供者之间过多的SLA违例。实验结果表明,策略在保证能效的同时,在减少SLA违例确保QoS方面也具有明显的效果。 3)提出了基于多数据中心的绿色高能效资源提供策略。数据中心的能效通常被多个动态因素影响,包括：能源成本、碳排放率、负载类型、CPU能效及冷却系统等,该策略将同时考虑以上因素研究跨越多个地理位置环境中的多数据中心的全局能效问题。首先建立了多数据中心的资源提供模型,将能耗制约的收益问题和碳排放(Carbon Footprint)问题形式化为QoS约束的收益函数和代价函数的多目标最优化模型,证明了该模型是NP-hard问题。针对该问题提出了绿色云优先的CMM、MCMP算法和收益优先的PMM、 MPMC算法,算法综合考虑了碳排放、能耗、收益和应用的QoS需求,目标是降低碳排放,增加收益,同时满足用户应用的QoS需求。执行应用阶段,在数据中心中利用提出的NDVS方法进一步优化能耗,求解了给定负载情况下单个数据中心功耗最小时CPU频率满足的条件,并求解了CPU的最优频率,证明了该频率下能耗达到局部极小。实验结果表明,策略不仅可以降低能耗成本,优化任务调度,而且还可以权衡碳足迹。 4)提出了基于遗传算法的虚拟机资源提供负载均衡策略。应用需求的多样性和节点资源的异构性不可避免地会导致资源提供过程中云计算节点的负载失衡问题,这极大地降低了云计算的整体资源提供效率。如何通过高效的负载均衡机制协调主机负载以提高资源利用率和系统性能是目前丞待解决的问题。针对这一问题,提出了基于负载均衡的虚拟机资源提供遗传算法VMPGALB, VMPGALB舍弃了传统二进制编码方法,采用了更适宜体现虚拟机提供特点的树型编码方案。制定选择策略时,采用基于适应度的比例选择策略和最优保存策略,该方法使得具有较小适应度的个体也有被选择的机会并直接保留最优个体至后代中。设计杂交算子时,通过对两个父代个体的交叉操作,并利用生成树方法,使VMPGALB具有更好的杂交性能。同时,为避免求解过程陷入局部最优,VMPGALB还按一定比例对产生的个体进行了变异操作。实验结果表明,比较传统遗传算法BGA、MOGA、启发式算法BFH和WLC,VMPGALB不仅遗传性能更优,虚拟机迁移次数更少,而且能以较快的收敛速度求解虚拟机提供的负载均衡方案。 5)提出了基于市场经济学模型的资源提供博弈策略。市场经济学模型可以通过均衡理论实现资源的优化配置,研究了以市场经济模型为基础的云计算资源提供机制,结合博弈论在资源管理领域的优势,首先,建立了非合作竞争市场的资源提供模型,提出了非合作博弈资源提供算法RPANCG,该算法以非合作博弈进行建模,RPANCG的目标是寻找使得各个资源提供者效用达到最优的Nash均衡解,证明了RPANCG算法可以产生唯一的Nash均衡。然后,在RPANCG算法满足效用相互最优的基础上,为了进一步增加集体收益,并满足效率与公平的约束,在非合作竞争市场的基础上提出了议价市场中的资源提供算法RPABG,该算法以议价博弈进行建模,RPABG的目标则是寻找Nash议价解。实验结果表明,RPANCG算法可以收敛到唯一的Nash均衡解,资源提供者的效用达到相互最优,整个资源提供趋于合理。而RPABG则在RPANCG算法的基础上进一步兼顾了资源分配的效率和公平性,并且能够提高资源提供者的整体效用,实现了Pareto改进,从而达到云资源的公平、合理和均衡的优化分配。本文的研究得到了国家自然科学基金项目(批准号：60970064,61272116),新世纪优秀人才支持计划项目(批准号：NCET-08-0806),教育部博士点基金项目(批准号：20120143110014)及湖北省高端人才引领培养计划项目的资助。
[Abstract]:As a new information processing model, Cloud Computing (Cloud Computing) technology has become a hot research focus in the information field. Cloud computing uses Virtualization as support technology to provide dynamic and scalable services to Internet users in a on-demand way. However, because of the large scale of cloud computing environment, resource management and division. With dynamic and scalable features, the energy consumption and resource availability of the cloud data center are the key factors affecting the performance of the cloud computing. In this paper, a new computing infrastructure, cloud computing technology, is used to study how to optimize the energy consumption and the optimal allocation of resources in cloud computing data centers. The energy consumption problem and its resources still exist a lot of problems to be solved. This paper focuses on the optimization methods of efficient resource provision in the cloud computing environment, including energy saving mechanism, load balance and market economy model. The main research work includes the following points:
1) the energy saving mechanism and resource optimization methods in cloud computing environment are studied systematically.
First, from the basic concept of cloud computing, it introduces the characteristics of cloud computing, the type of service and the level of service. Secondly, it focuses on the energy saving optimization strategy in cloud computing, analyzes and compares the application environment and advantages and disadvantages of the strategy. Then, it further studies the source supply technology in the cloud computing, and advances the current optimization strategy in this field. The classification and comparison are carried out. Finally, the cloud experiment environment CloudSim is introduced, and its resource providing mechanism is experimentally analyzed.
2) a virtual machine resource providing strategy based on energy and SLA equilibrium is proposed.
In view of the dynamic change characteristics of application requirements in the cloud computing environment, a virtual machine adaptive deployment algorithm RLWR based on strong local weighted regression is proposed. RLWR can dynamically decide the overloading time of the host based on the resource occupying historical information embodied in the applied load. After the overloaded main machine is detected, the virtual machine migration with the best migration cycle is proposed. The migration selection algorithm MPM and the least migration algorithm MNM are used to select the migration virtual machine, and then the optimal deployment of the migrated virtual machine is redeployed with the descending optimal adaptive heuristic algorithm PBFDH based on the power consumption. The adaptive deployment strategy compares the static threshold algorithm STH, MPA and DVFS, which can not only dynamically deploy the virtual machine to less. On the physical host, it closes the idle host, improves the energy efficiency, and realizes high reliable QoS service delivery through the load prediction of the host resource, avoiding excessive SLA violation between the user and the resource provider. The experimental results show that the strategy also has obvious effect in reducing the SLA violation to ensure the QoS while guaranteeing the energy efficiency.
3) a green energy efficient resource delivery strategy based on multi data center is proposed.
The energy efficiency of the data center is usually influenced by several dynamic factors, including energy cost, carbon emission rate, load type, CPU energy efficiency and cooling system. The strategy will consider the above factors at the same time to study the global energy efficiency problems of multi data centers across multiple geographic environments. The problem of energy consumption and carbon emission (Carbon Footprint) is formalized into a multi-objective optimization model of revenue function and cost function of QoS constraints. It is proved that the model is a NP-hard problem. The green cloud priority CMM, MCMP algorithm and revenue priority PMM, MPMC algorithm are proposed for the problem, and the carbon scheduling algorithm is taken into consideration. The QoS demand for energy consumption, revenue and application is designed to reduce carbon emissions, increase revenue and meet the QoS requirements of user applications. In the application phase, the proposed NDVS method is used to further optimize energy consumption in the data center, and the conditions for the minimum power consumption of a single data center in the case of a given load are solved, and the C is solved, and the C is solved. The optimal frequency of PU shows that the energy consumption at this frequency reaches a local minimum. The experimental results show that the strategy not only reduces the cost of energy consumption, optimizes the task scheduling, but also can balance the carbon footprint.
4) proposes a load balancing strategy based on genetic algorithm for virtual machine resources.
The diversity of application requirements and the heterogeneity of node resources inevitably lead to the problem of load imbalance of cloud computing nodes in the process of resource provision, which greatly reduces the overall resource efficiency of cloud computing. How to coordinate host load through efficient load balancing mechanism to improve resource utilization and system performance is now the prime minister To solve this problem, a genetic algorithm VMPGALB is provided for virtual machine resources based on load balancing, and VMPGALB abandoning the traditional binary coding method and adopting a tree coding scheme which is more suitable to provide the features of virtual machines. In this method, the method makes the individuals with smaller fitness have the opportunity to be selected and retain the optimal individual directly to the offspring. When designing the crossover operator, the cross operation of two parent individuals and the spanning tree method make VMPGALB have better hybrid performance. At the same time, in order to avoid the solution process falling into local optimal, VMPGA LB also performs the mutation operation on a certain proportion. The experimental results show that compared with traditional genetic algorithms BGA, MOGA, heuristic algorithm BFH and WLC, VMPGALB not only has better genetic performance and less migration times, but also can solve the load balancing scheme provided by virtual machine at a faster rate of convergence.
5) put forward a game strategy of resource supply based on market economics model.
The market economy model can optimize the allocation of resources by equilibrium theory, study the mechanism of providing cloud computing resources based on the market economy model, combine the advantages of game theory in the field of resource management. First, it establishes a resource supply model for non cooperative competition market, and proposes an algorithm RPANCG for non cooperative game resource provision. The algorithm is modeled by non cooperative game. The goal of RPANCG is to find the Nash equilibrium solution that makes the utility of various resource providers reach the best. It is proved that the RPANCG algorithm can produce the unique Nash equilibrium. Then, on the basis of the RPANCG algorithm satisfying the mutual optimal utility, the algorithm can increase the collective income and meet the efficiency and fairness. Constraint, based on the non cooperative competitive market, the resource provision algorithm RPABG in the bargaining market is proposed. The algorithm is modeled by the bargaining game, and the goal of RPABG is to find the Nash bargaining solution. The experimental results show that the RPANCG algorithm converges to the only Nash equilibrium solution, the utility of the source provider reaches each other optimal, and the whole resource is proposed. The supply tends to be reasonable, and RPABG is based on the efficiency and fairness of resource allocation on the basis of RPANCG algorithm, and can improve the overall utility of resource providers and realize the improvement of Pareto, thus achieving fair, reasonable and balanced allocation of cloud resources.
The research obtained from the National Natural Science Foundation Project (approval number: 6097006461272116), the new century excellent talent support program (approval number: NCET-08-0806), the PhD fund project of the Ministry of Education (approval number: 20120143110014) and the financing plan project of the high-end talents in Hubei province.
【学位授予单位】：武汉理工大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP3;F205

【参考文献】