云环境下虚拟网络动态演化管理技术研究

发布时间：2018-02-24 12:52

本文关键词： 云计算云应用服务质量目标虚拟网络虚拟网络重配置概率推理　出处：《大连理工大学》2015年博士论文　论文类型：学位论文

【摘要】：网络虚拟化技术将网络资源层与物理层解耦,以融合分布异构网络资源的方式为上层应用屏蔽网络复杂性问题,提升网络配置管理灵活性。云计算与网络虚拟化技术的结合使得实现更敏捷、高效的网络资源管理成为可能。然而,由于云应用的动态性,以及缺少业务感知的主动、自适应虚拟网络管理能力,云环境下虚拟网络资源利用率仍然不高,网络碎片化等管理问题日趋严峻。虚拟网络动态适应云应用业务需求能力不足,不但不能提升云应用灵活性、敏捷性和资源利用率,反而会因为额外构建的虚拟网络层增加云基础设施的复杂度,从而带来潜在的稳定性和性能风险,并造成更大的资源浪费。随着面向未来的“端—网—云”模式的出现和快速演进,网络主动发现、识别云应用的动态性并持续演化适应动态性的能力不足,与快速增长的网络互联终端数量之间的矛盾日益突出。目前,虽然该领域已经出现一些相关技术研究成果,但仍然存在虚拟化成本过高,虚拟节点迁移有效性难以判断的管理复杂性问题；基于主机内部虚拟化技术构建的网络碎片化及资源利用率低；虚拟网络资源与业务目标不能动态自适应,以及现有虚拟网络重配置算法执行效率不高问题仍待突破解决。本文分析了当前面临的问题与挑战,深入研究了相关技术现状并针对性地展开技术研究并提出了基于信念-愿望-意图(Belief-Desire-Intention, BDI)模型的虚拟资源自治管理框架。以此为支撑,利用图论分裂聚类理论、贝叶斯信念网络(Bayesian Belief Network, BBN)概率推理、线性规划等理论模型解决了云应用服务质量目标感知的虚拟网络动态演化管理相关问题,提升了现有虚拟网络资源调度以及虚拟网络重配置算法执行效率。最后基于理论研究成果,设计实现了面向云应用的敏捷虚拟网络管理平台系统,通过测试验证了技术有效性。本文的主要工作可归纳为以下几点：(1)设计了一种基于信念-愿望-意图模型和多智能体系统的虚拟资源自治管理体系结构,提出了以云应用为核心的虚拟资源自治管理框架设计方案,通过将自治控制功能与云应用本身解耦的策略简化了管理复杂度,和网络接入云平台配置、管理的难度,解决了网络虚拟化损耗过高,以及虚拟资源节点迁移有效性判定的管理问题,为实现业务目标感知的虚拟网络动态演化管理提供框架支撑。(2)提出了基于图论分裂聚类算法的云应用模型容器内虚拟网络资源自适应调度策略,实现了通过判断虚拟网络带宽资源使用率对基于云应用容器模型描述的虚拟网络节点调度,优化其网络带宽资源分配策略的方法,缓解了网络碎片化及性能干扰问题,提升了物理网络资源利用率。(3)提出了服务质量目标感知的虚拟网络风险分析、定位方法,基于贝叶斯信念网络设计了对由于异构网络环境计算通信资源的动态、多元性导致的虚拟网络性能不确定性,进行因果关系建模(Causal-Effect Relation Modeling)的技术方案,并利用预先给定的服务质量目标,主动触发贝叶斯网络最大概率解释MPE (Most Probability Explanation)概率推理,判断虚拟网络超配／不足状态的策略,实现了自动化风险分析定位,以及随需动态演化式的虚拟网络重配置请求生成,大幅降低了人工运维复杂度及工作量。(4)提出了虚拟节点／虚拟链路迁移与置换策略相结合的虚拟网络重配置方法,定义了通过虚拟网络重配置率指标,并基于此来衡量虚拟网络重配置请求中,资源提升需求满足程度的策略,通过利用线性规划理论对目标问题域建模,设计了虚拟网络重配置算法,进一步提升了现有重配置算法执行效率。(5)设计实现了具有随需交付虚拟网络资源能力的敏捷虚拟网络管理架构,能够通过动态感知云应用为核心的服务质量目标,来实现虚拟网络资源调度,以及拓扑结构持续演化,为“端—网—云”架构系统提供基础管理平台级支撑。
[Abstract]:Network virtualization technology cyber source layer and physical layer decoupling, application layer shielding network complexity to integration of distributed heterogeneous cyber source way, enhance the flexibility of network configuration management. With the combination of cloud computing and network virtualization technology enables more agile, efficient management of cyber source as possible. However, due to the dynamic nature of cloud applications and the lack of service awareness, initiative, adaptive virtual network management capabilities, the utilization rate of the virtual cyber source cloud environment is still not high, the network fragmentation and other management issues are becoming increasingly serious. The dynamic virtual network to adapt to the cloud application service demand ability is insufficient, not only can improve the utilization rate of cloud application flexibility, agility and resources, but because of the construction of additional virtual network layer increases the complexity of the cloud infrastructure, so as to bring stability and performance of the potential risks, and cause more. The source of waste. As a future oriented "end net cloud" model and the rapid evolution of network, active discovery, dynamic cloud application identification and continued evolution of the ability to adapt to dynamic problems, with the contradiction between the rapid growth of the number of Internet terminals have become increasingly prominent. At present, although the field has some the related research achievement, but there is still a virtual high cost and complexity of management effectiveness is difficult to determine the virtual node migration; the low utilization rate of the main internal construction of virtualization technology based on network fragmentation and resources; the virtual cyber source and business objectives can not be dynamically adaptive, as well as the existing virtual network reconfiguration algorithm execution efficiency is not yet the problem of high breakthrough in solving. This article analyzes the problems and challenges facing, in-depth study of the related technical status and pointed out the research and technology The belief desire intention based on (Belief-Desire-Intention, BDI) autonomous virtual resource management framework model. In order to support, split by graph clustering theory, Bayesian belief network (Bayesian Belief Network, BBN) probabilistic reasoning, linear programming model to solve the related problems of the evolution of dynamic management of virtual network cloud application service quality objectives perception, to enhance the existing virtual cyber source scheduling and virtual network reconfiguration algorithm efficiency. Finally based on the theoretical research, design and Implementation for the cloud applications agile virtual network management platform system, through the test to verify the effectiveness of the technology. The main work of this paper can be summarized as follows: (1) design. Based on the belief desire intention model and multi-agent system for autonomous virtual resource management system structure, puts forward the application as the core of the virtual cloud Autonomous resource management framework design scheme, the autonomous control function and cloud application itself decoupling strategy simplifies the complexity of management, network access and cloud platform configuration, management difficulties, solve the network virtualization and resource loss is too high, the virtual node migration management validity judgement, provide a framework to support management for dynamic virtual network to achieve the business target perception evolution. (2) proposed a cloud application model graph clustering algorithm in container splitting virtual cyber source adaptive scheduling strategy based on the judgment through the implementation of virtual network band width resource utilization rate of virtual network node scheduling description model based on cloud application container, the optimization method of network bandwidth resource allocation strategy. To ease the interference and performance of network fragmentation problems, enhance the physical and cyber source utilization. (3) proposed the service quality perception of virtual target The positioning method of network analysis, risk, Bayesian belief network design due to the heterogeneous network environment of computation and communication resources based on dynamic virtual network performance diversity leads to uncertainty of causal modeling (Causal-Effect Relation Modeling) technology, and use the quality of service given in advance, active trigger the maximum probability of Bayesian network interpretation MPE (Most Probability Explanation) probabilistic reasoning, judgment of virtual network with super / lack of state strategy, realize automatic positioning and risk analysis, virtual network on-demand dynamic evolutionary reconfiguration request generation, greatly reduces the complexity of the operation and maintenance manual and workload. (4) proposed a virtual network reconfiguration method combining virtual virtual node / link migration and replacement strategy, defines the index through the virtual network reconfiguration, and based on this measure The virtual network reconfiguration request, to meet the needs of resources to enhance the degree of strategy, by using linear programming theory to target problem domain modeling, virtual network reconfiguration algorithm design, to further improve the existing reconfiguration algorithm efficiency. (5) designed an on-demand delivery of virtual cyber source capability of agile virtual network management architecture that can service the quality target through the dynamic perception of cloud application as the core, to realize the virtual cyber source scheduling, and the topological structure of continuous evolution, provide basic management platform support for the "end - Network - cloud architecture system.

【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TP393.01

【相似文献】