移动云服务架构设计与计算卸载策略研究

发布时间：2018-08-05 20:32

【摘要】：伴随互联网的高速发展以及基于“云”的移动应用的爆发式增长,数据流量急剧升高,移动通信网络需要寻求新的发展方向以满足更高的服务需求,但无论未来网络如何演进,“云”的特征必不可少。云计算与移动通信网络相结合,将有利于形成通信、计算、存储“三位一体”的网络形式,利用通信网络承载计算任务,利用计算技术来解决通信问题。因此新型的计算与通信相融合的云服务网络架构急需被提出并具备深远意义。计算卸载是移动云计算的基本原理与主要服务交付模式,有利于用户突破终端匮乏的物理资源的限制,实现复杂的云服务和更优的用户体验。计算卸载主要具备的特性为交互性和置换性,用户与云端在信息交互的过程中,通过消耗部分通信能耗与带宽资源置换云端强大的存储/计算资源。是否进行计算卸载要考虑时间成本、能耗成本等多个因素,因此提出合理的计算卸载策略有利于用户花销的最小化和网络利用的最优化。针对上述两个问题,本文首先对支撑移动云服务的整体网络架构进行重构与创新,然后综合考虑任务信息,云端信息和通信网络信息,提出适用于该架构的多站点机会计算卸载策略。本文主要的研究工作与创新性如下:(1)基于计算和通信相融合的思想,本文设计了分层次分布式的移动云服务架构,基于传统的核心云服务,融入新兴的边缘云服务,并创新性的提出设备云服务。利用边缘云将云服务拉近到接入侧,减少通信时延并缓解主干链路压力;利用设备云将云服务进一步融入到用户端,实现终端的协同计算。同时本文还设计了基于软件定义网络(Software-Defined Networking,SDN)技术的协同控制云服务系统,有效的缓解分散的云资源带来的管理难题。(2)基于分层次分布式部署的云服务架构,本文研究了基于SDN局部协同控制器指导的多站点计算卸载服务交付流程,分析了移动环境下计算卸载服务具备的机会计算特性。提出了移动环境下“终端-云端”有效连通概率的计算模型。最后通过Matlab对所提计算模型进行仿真与验证,证明了该模型的有效性。(3)基于多站点机会计算卸载服务流程,本文综合考虑计算任务信息、云端信息和网络信道信息,设计了多站点计算卸载策略。本文用计算量和传输量定义任务状态、用有效连通概率衡量云端状态、用Gilbert-Elliott(GE)信道模型拟合无线信道状态,以最小化系统花销(包括时间花销、能耗花销和惩罚花销)为目标将计算卸载问题转化为最优化问题,建立基于带有约束的马尔科夫决策的数学模型,并利用带有Q学习的向后迭代算法求解计算卸载的最优策略。最后,通过Matlab仿真对比其他三种策略,结果证明了算法的优越性,可以有效的减少用户花销。
[Abstract]:With the rapid development of the Internet and the explosive growth of mobile applications based on "cloud", the data flow increases sharply. Mobile communication networks need to seek new development direction to meet higher service demand, but no matter how the future network evolves, The features of the cloud are essential. The combination of cloud computing and mobile communication network will facilitate the formation of communication, computing, storage "trinity" network form, using the communication network to carry computing tasks, using computing technology to solve communication problems. Therefore, a new cloud service network architecture which combines computing and communication needs to be proposed and has profound significance. Computing uninstall is the basic principle and main service delivery mode of mobile cloud computing, which is helpful for users to break through the limitation of physical resources, and to realize complex cloud services and better user experience. The main characteristics of computing and uninstalling are interactivity and permutation. In the process of information exchange between user and cloud, the powerful storage / computing resources of cloud are replaced by consuming part of communication energy and bandwidth resources. Time cost, energy consumption cost and other factors should be taken into account in the calculation of uninstall. Therefore, a reasonable calculation and unloading strategy is proposed to minimize the cost of users and optimize the utilization of the network. Aiming at the above two problems, this paper first reconstructs and innovates the overall network architecture that supports mobile cloud services, and then synthetically considers task information, cloud information and communication network information. A multi-site opportunity computing uninstall strategy for this architecture is proposed. The main research work and innovation are as follows: (1) based on the integration of computing and communication, this paper designs a hierarchical distributed mobile cloud service architecture, which is based on the traditional core cloud services and integrates into the emerging edge cloud services. And innovative device cloud services. The edge cloud is used to bring the cloud service closer to the access side to reduce the communication delay and relieve the backbone link pressure. The device cloud is used to further integrate the cloud service into the client to realize the cooperative computing of the terminal. At the same time, this paper also designs a collaborative control cloud service system based on Software-Defined networking (SDN) technology, which can effectively alleviate the management problems caused by distributed cloud resources. (2) the cloud service architecture based on hierarchical distributed deployment. In this paper, the delivery process of multi-site computing unload service based on SDN local cooperative controller is studied, and the opportunistic computing characteristics of computing uninstall service in mobile environment are analyzed. A model for computing the effective connectivity probability of "terminal-cloud" in mobile environment is presented. Finally, the proposed model is simulated and verified by Matlab, which proves the validity of the model. (3) based on the multi-site opportunity to calculate the unloading service flow, this paper comprehensively considers computing task information, cloud information and network channel information. A multi-site calculation and uninstall strategy is designed. In this paper, the task state is defined by computation and transmission, cloud state is measured by effective connectivity probability, and wireless channel state is fitted by Gilbert-Elliott (GE) channel model to minimize system cost (including time cost). Energy consumption and penalty cost) as the goal, the computational unload problem is transformed into an optimization problem, a mathematical model based on constrained Markov decision is established, and a backward iterative algorithm with Q learning is used to solve the optimal strategy for calculating unload. Finally, compared with other three strategies through Matlab simulation, the results show that the superiority of the algorithm can effectively reduce the cost of users.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.5;TP393.09

【相似文献】