高可靠云服务供应关键技术研究

发布时间：2018-12-08 16:43

【摘要】：云计算以其计算能力按需提供、基于虚拟机化集群技术的资源共享、快速弹性扩展与收缩等特性受到学术界和工业界的广泛关注。然而,云计算环境下,由于资源的大规模性、节点失效事件发生的不确定性、竞价型虚拟资源获取的非持续性,导致云服务可靠性急剧下降。为此,本文以云虚拟资源供应和云应用服务供应为主要研究对象,以高可靠、低代价为研究目标,对两类云服务可靠供应中面临的若干关键问题进行了深入研究,取得了如下成果：1.针对当前可靠云虚拟资源供应研究中,由于忽略节点可靠性和数据中心网络拓扑特征,导致已有基于关键点存储/恢复的可靠性保障策略难以达到最优效果,和资源占用率高的问题,本文首先借鉴可靠性理论,通过设计特征融合因子对数据中心节点的可靠性进行建模,并将虚拟机关键点存储成本转换为数据中心资源消耗成本,然后针对云服务和胖树型数据中心网络特征,提出了一种共享式虚拟机关键点镜像文件生成与调度策略。最后,在基于Cloudsim构建的云仿真平台FTCloudsim上进行了大量仿真实验,实验结果表明,与其它方法相比,所提方法能保证可靠性的同时,降低对数据中心网络资源和存储资源的消耗。2.针对当前可靠云虚拟资源供应研究中,已有虚拟机冗余策略在失效恢复过程中占用大量网络资源的问题,提出一种数据中心资源占用特征和数据中心网络拓扑特征感知的虚拟机冗余放置和任务重调度方法,该方法首先基于聚集特征设计物理机选择算法,其次利用数据中心结构特征设计启发式算法,计算最优虚拟机放置策略。最后,将重调度问题建模为二分图最小权最优匹配问题,并设计匹配算法,降低实际失效恢复时间,提高网络资源利用率。为了验证本方法的有效性,我们在FTCloudsim上进行仿真实验,实验结果表明所提方法能保障云虚拟资源供应可靠性的同时,显著提升数据中心网络资源利用率。3.针对当前可靠云应用服务供应研究中,由于缺乏对竞价型资源价格特征准确认知和对任务紧急度的忽略,难以实现可靠性和运营成本均衡的问题,提出一种自适应竞价型虚拟资源动态租赁与可靠性保障方法,该方法首先利用竞价型虚拟资源的历史价格和任务的紧急度特征,设计竞价策略,其次,以状态存储决策周期为基线,基于历史价格特征,对潜在虚拟机状态保存策略的代价进行评估,并选择代价最小的策略作为目标策略。为了验证所提方法的有效性,我们在FTCloudsim上进行了大量仿真实验。实验仿真对比表明所提出的方法能够保障云应用服务可靠性,并降低应用服务供应商的运营成本。
[Abstract]:Cloud computing has attracted much attention from academia and industry because of its ability to provide on demand, resource sharing based on virtual machine cluster technology, rapid elastic expansion and contraction. However, in the cloud computing environment, due to the large-scale nature of resources, the uncertainty of node failure events, and the non-sustainability of bidding virtual resources acquisition, the reliability of cloud services is reduced sharply. For this reason, this paper takes cloud virtual resource supply and cloud application service supply as the main research object, taking high reliability and low cost as the research target, and makes a deep research on some key problems in the two kinds of cloud service reliability supply. The following results have been achieved: 1. In the current research of reliable cloud virtual resource supply, the reliability of nodes and the topology of data center network are ignored, so it is difficult to achieve the optimal effect of the existing reliability guarantee strategy based on key point storage / recovery. And the problem of high resource utilization, this paper first uses the reliability theory for reference, models the reliability of the data center node through the design feature fusion factor, and converts the storage cost of the virtual machine key point into the resource consumption cost of the data center. Then, according to the network features of cloud services and fat-tree data centers, a shared virtual machine key point image file generation and scheduling strategy is proposed. Finally, a large number of simulation experiments are carried out on the cloud simulation platform FTCloudsim based on Cloudsim. The experimental results show that compared with other methods, the proposed method can guarantee reliability at the same time. Reduce the consumption of data center network resources and storage resources. 2. In the current research of reliable cloud virtual resource supply, the redundant strategy of virtual machine occupies a lot of network resources in the process of failure recovery. A virtual machine redundancy placement and task rescheduling method based on data center resource occupation feature and data center network topology feature perception is proposed. Firstly, a physical machine selection algorithm based on aggregation feature is designed. Secondly, a heuristic algorithm is designed based on the structural features of the data center to calculate the optimal virtual machine placement strategy. Finally, the rescheduling problem is modeled as the bipartite graph minimum weight optimal matching problem, and a matching algorithm is designed to reduce the actual failure recovery time and improve the network resource utilization. In order to verify the effectiveness of this method, we have carried out simulation experiments on FTCloudsim. The experimental results show that the proposed method can guarantee the reliability of cloud virtual resource supply and improve the utilization ratio of data center network resource significantly. In the current research of reliable cloud application service supply, it is difficult to achieve reliability and operation cost balance due to the lack of accurate cognition of price characteristics of bidding resources and neglect of task urgency. In this paper, an adaptive bidding virtual resource dynamic lease and reliability guarantee method is proposed. Firstly, a bidding strategy is designed based on the historical price of the bidding virtual resource and the urgency of the task. Taking the state storage decision cycle as the baseline and based on the historical price characteristics, this paper evaluates the cost of the state preservation strategy of the potential virtual machine, and selects the strategy with the lowest cost as the target strategy. In order to verify the effectiveness of the proposed method, we have carried out a large number of simulation experiments on FTCloudsim. The simulation results show that the proposed method can guarantee the reliability of cloud application services and reduce the operating costs of application service providers.
【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TP393.09

【相似文献】