可靠性驱动的移动云资源管理随机模型和优化方法

Mobile Cloud Computing system is undergoing a dramatic change with the Industrial revolution of Mobile Internet. The resource information of mobile cloud computing system with ‘Reliability as a server’ must be shared in real-time, optimization regrouping resource can reliably and on-demand meet the service requirements of users, which are vast and differentiation. Optimization of resource management to achieve the best match between the allocation of resources and the service demand and improvement the utilization of resources is an effective way...This proposal addresses the contradiction between the limited resources and the increasing demand, which partially answers the three key scientific questions in the study of the resource management on mobile cloud computing systems. First, an optimization scheduling of fault-tolerant resource using extremal graph theory and an exact mathematical model of resource management based on the reliability stochastic model are put forward, which provide a general method and solution to describe the mobile cloud resource systems by modeling. Second, by Comprehensive application of the fuzzy theory and stochastic optimization theory, we propose a resource management model in accordance with service demand clustering, which gives a modeling method for variety service requirements description. Finally, we prove that the cost and the reliability requirements is generalized Nash equilibrium using the game theory, an optimization resource configuration scheme by multi-objective constraints is also designed...The expected results can provide the innovative ideas and real solutions for mobile cloud computing theory, as well as some good technical support for the widespread use of mobile cloud computing in the national economic and social development.

移动云计算系统正在经历一场移动互联网产业变革带来的巨大变化。 “可靠性即服务”的移动云要求实时共享资源信息，可靠的优化组合服务资源以满足用户海量的、差异化的服务需求。优化资源管理、实现资源分配与服务需求的最佳匹配、提高资源利用率是一种有效途径。本项目针对移动云资源管理中的3个关键科学问题开展研究：首先，运用极值图论设计容错资源的优化调度策略，并基于可靠性随机模型提出资源管理的精确模型，为模型化描述移动云资源系统提供了一般性方法和通用求解技术；其次，综合应用模糊理论和随机优化理论，构建服务需求聚类的资源管理模型，为多样的服务需求描述提供了一种模型化方法；最后，基于博弈论理论，证明了成本约束和可靠性需求之间的广义纳什均衡，并设计了多目标约束的资源配置与性能优化方案。预期成果能够为移动云理论进一步完善提供创新性思路和现实解决方案，从而为移动云在国民经济和社会发展中的广泛应用提供有益

研究可靠性驱动的移动云资源随机管理模型和优化方法。取得的主要研究成果包括：依据随机更新回报理论，结合系统故障概率分布函数和累积分布函数，设计了一种基于分级检查点技术的容错算法，对分级检查点设置频率进行了动态确定，基于超图覆盖的分级检查点存储策略可以有效的提高云云服务质量；结合预测机制，动态的云副本数确定算法采用指数平滑预测机制预测文件未来的访问量，并结合文件当前的与未来的访问量计算出文件的副本数，有效提高了文件的副本的访问效率；提出了网络系统可靠性稳定度评估方法，对系统组件的风险增长因子和网络系统可靠性稳定度进行了分析，可以准确的发现网络系统组件的脆弱环节，从而实现对云系统故障组件的快速定位；设计了一种云资源的新的抽象算法以及新的多目标能效优化函数，进而提出了一种基于全局优化的粒子群算法，实现了对云资源分配的多目标优化，减少了任务完成总时间和总开销；对基于影响空间的局部异常因子算法（INFLO）进行了增量式改进，设计了虚拟机常检测检测模型，对新采集的虚拟机实例进行增量式地异常检测，从而提高异常检测效率；针对资源分配效率低下问题，通过对任务需求和与资源属性聚类，设计了用户任务与云资源双边聚类模型和任务资源严格匹配策略，提高了资源匹配双方的综合满意度；针对移动终端在移动时的网络切换选择导致用户竞争从而降低数据传输速率问题，引入预测切换原理，设计了候选网络预测接入模型和基于潜在博弈理论的计算任务卸载策略，比传统的典型卸载方案大概优化了10.3%。