【作者】 王宁；

【导师】 杨扬；

【作者基本信息】 北京科技大学 ， 通信与信息系统， 2015， 博士

【摘要】 云计算拓宽了用户可访问资源的范围,构建了一种以服务形式弹性、按需获取各类信息资源的架构,成为当前多种信息应用部署和扩展的坚实基础环境。云计算强化了面向服务计算的理念,其处理的对象被视为来自用户的各类数据,满足用户需求和实现资源利用效率的提高是云计算系统处理和优化管理各类数据的出发点。受到网络环境、资源分布情况、服务形态,以及模式等因素的影响,数据的优化管理和高效调度问题面临诸多挑战。在总结云计算中数据管理和优化调度成果的基础上,本文分别从数据的利用效率、各类环境因素的影响和物理节点失效等角度,重点考虑了运营成本最优化的问题,从数据管理和优化调度的角度进行了较深入的探讨,给出了基于热度的数据部署策略、部署节点性能的综合度量方法和应对节点失效的数据修复机制,并通过仿真和实验的方式进行了验证。成果将有利于云计算服务提供商在保证服务质量的前提下,实现运营成本的优化,具有较大的理论和现实意义。本文的主要研究工作和创新成果如下。1)提出了基于热度的数据块优化存储策略,实现了服务成本最优的数据块部署。针对常采用的对数据进行分块和分布式冗余存储的机制,引入数据块热度(用户对数据块的兴趣需求)的概念,给出了不同热度数据块应具备的最优冗余存储个数；进一步考虑被用来部署服务数据节点的成本差异,得到了一种满足用户需求的存储数据块的最小服务成本策略；通过引入调整因子,得到可针对不同负载情况的自适应数据块最小服务成本策略。2)提出了一种刻画可用节点服务性能的综合度量方法,能实现多指标影响下的数据块优化调度部署。针对数据块的业务特性,从节点成本、负载率、带宽、网络延迟和任务的计算复杂度等角度考虑了备选部署节点的综合效用值,给出一种综合效用评价指标,并在此基础上,得到了服务质量优化的任务调度模型和策略。3)给出了一种能够应对节点失效的数据恢复方法,保证在服务总成本最小的情况下数据的高效恢复和使用。为应对云计算系统中的节点失效带来的数据丢失和错误,我们提出了一种基于差异对待的数据重新备份和部署机制,考虑数据块的热度,设计了选择性恢复算法和成本矩阵调整因子,实现了数据的弹性恢复,并在不改变存储数据块最小服务总成本的情况下,提高了用户对内容访问的服务性能。更多还原

【Abstract】 Cloud computing has expanded the scope of the accessed resources. It supplies a solid base environment for multiple information applications and extensions by establishing an architecture in which information resources can be fetched elastically and on-demand in the form of services. Cloud computing is the reinforcement of the Service-orient Architecture (SOA). It processes various kinds of data that come from customers and aims at meeting their needs and increasing the resource utilization rate. To realize this goal, many cloud computing systems and their optimizing strategies have been proposed. However, considering the influences of network status, resources distribution, service forms and patterns, there are still a lot of challenges in the optimal management of data and efficient scheduling.Based on the existing work in the data management and optimal scheduling, this dissertation goes deeply into the optimization of operating costs. Synthesizing a variety of system and environmental factors, as well as the failures of physical node, a data deployment strategy is proposed according to block-popularity. Furthermore, a method of a new node performance measurement and a mechanism of failure node repair are proposed. The strategies are verified by simulation and experiments. The results show that they can help service providers to optimize their operating costs without reducing the Quality of Service (QoS). The study of these issues has important industrial and academic value.The main contributions of this dissertation are as follows.1) A data block storage strategy based on block-popularity is proposed, targeting at block deployment with the optimization of the service cost. In the proposed strategy, data blocks with different popularities will have different optimal replicas. The differences in the costs of service nodes are further considered and a minimum service cost strategy that meets consumers’needs is proposed. By introducing adjustment factor and considering system load, an adaptive strategy of minimum cost of storing blocks is provided.2) A measurement method of node performance is proposed, which can optimally deploy the data blocks with multiple parameters. By taking node business characterization into consideration, the synthesized utility values of the alternative node are considered regarding to the node cost, load rate, bandwidth, network delay and computation complexity of the tasks. Evaluation criterion of synthesized utility is given. Basing on it the task scheduling model and strategy with optimal service quality are proposed.3) To guarantee efficient data recovery and utility with minimum service cost, we proposed a data recovery method to deal with node failure. In this method, a selective data recovery algorithm and a date redeployment strategy are presented. Considering block-popularity, a cost matrix adjustment factor is also provided to recover lost data elastically and improve the service performance of content access without changing the minimum total service costs.更多还原