【作者】 孙晓川；

【导师】 刘韵洁；

【作者基本信息】 北京邮电大学 ， 通信与信息系统， 2013， 博士

【摘要】 互联网经过几十年的发展,已成为当今世界上覆盖范围最广、规模最大、信息资源最丰富的全球信息基础设施。然而,随着网络规模的不断扩大以及新兴网络通信技术和互联网业务的发展,传统互联网的“僵化”问题逐渐凸显,已经不能适应未来网络的可持续发展,主要表现在可扩展性、可管可控可测性、安全性、可移动性、绿色节能等方面。为此,国内外积极开展未来网络理论的研究工作,涉及未来网络体系结构、网络虚拟化技术、网络测量与预测等研究领域。为了从根本上解决当前互联网面临的问题,本论文从网络虚拟化资源管理架构、虚拟化资源分配机制和网络流量预测算法三个方面对未来网络虚拟化资源管理进行深入的研究,主要研究内容及创新包括：(1)针对虚拟化资源管理架构可扩展性不强、自治性差、缺乏测量感知能力等问题,本文提出了一种基于预测的智能化分布式虚拟化资源管理架构(IDP-VRMA),采用分层分域的管理策略,实现了对本管理域的“集中式管理和分布式控制”。每个管理域中的管理系统负责接收虚网请求、注册、虚网拓扑分割以及与其他管理域的管理系统联系完成跨域的虚拟化资源分配；集成在底层网络节点内的智能控制器可以自治地执行资源分配、故障修复、威胁处理等；知识处理部分实时监控整个管理域内的网络运行情况,挖掘出有效的知识信息,从而为管理系统和智能控制器执行相关的管理和控制策略提供必要的依据。该架构有效地解决可扩展性问题,自治地实现虚拟化资源配置、网络故障和威胁处理,并且具备测量感知能力,大大提高了虚拟化资源分配的有效性和可靠性。(2)在IDP-VRMA架构中,要实时地为动态到达的属于不同SPs的众多虚网请求分配底层网络资源,这些虚网请求有着不同的拓扑结构,以及对多种异构资源的需求和约束,使得多个VNs之间以及与InPs之间存在着复杂的关系,即多个VNs同时竞争底层网络多个组件中的多种资源。为了建模该复杂关系,本文提出了一种基于多维非合作博弈的虚拟化资源分配机制,通过效用函数、价格函数、拥塞函数三部分建模了多维非合作博弈模型中各虚网的总支付函数,并证明了该模型的多维Nash均衡的存在性和唯一性。仿真结果表明,该模型在博弈的各个领域内能够达到多维Nash均衡,有效地抑制VNs的自私行为,公平、合理、高效地分配网络虚拟化资源。(3) IDP-VRMA虚拟化资源管理架构下,需要一种稳定、快速并具有良好预测性能的网络流量预测算法。针对这一问题,本文提出了一种具有环形反馈动态池结构的确定回声状态网络(ALR),并应用到网络流量预测中。通过在最简单环形动态池中相邻的神经元之间引入邻接反馈,构建了该模型的动态池,并且设置其输入权值绝对值和内部权值相等,都为r,因此仅仅需要调节一个自由参数,极大地简化了回声状态网络的结构。通过该模型的状态更新方程和输出方程,进一步证明了其良好的记忆能力。仿真结果表明,该模型能够很好地刻画网络流量的自相似性,具有良好的预测性能。更多还原

【Abstract】 After decades of development, Internet has become a globe information infrastructure with the widest coverage, largest scale and most abundant infor-mation resource in today’s world. With the continuous expansion of network scale, as well as the development of new network communication technolo-gy and Internet business, the "ossification" problems of the traditional Internet have become prominent gradually. It has been unable to meet the sustainable development of future network, mainly in scalability, controllability, manage-ability, measurability, security, mobility and green energy. To this end, the research for future network theory has been actively carried out at home and abroad, involved in the research areas of the future network architecture, net-work virtualization technology, network measurement and prediction.In order to fundamentally solve the problems of the current Internet, this paper studies the virtual resource management for future network from the as-pects of network virtual resource management architecture, virtual resource allocation mechanism and network traffic prediction algorithm. Major research and innovation include:(1) For poor Scalability and autonomy of current network virtual resource management architecture, as well as the lack of measurement&perception, this paper proposes a prediction-based intelligent and distributed virtual resource management architecture (IDP-VRMA). The architecture, using the manage-ment strategy based on multi-layer and multi-domain, realizes the centralized management and distributed control in the local management domain. In each management domain, the management system is responsible for receiving vir-tual network requests, registration, virtual network partition, and contacting with management systems in other management domains to complete cross-domain virtual resource allocation; the intelligent controllers integrated in the substrate network nodes can autonomously perform virtual resource allocation, fault repair, threat handling; the knowledge processing element can monitor the network operation, mine effective knowledge, contributed to perform the corresponding management and control strategies for management system and intelligent controllers. The IDP-VRMA can effectively solve the scalability problem, autonomously implement virtual resource configuration and network failure&threats treatment, have network measurement and perception, greatly improve the validity and reliability of the virtual resource allocation.(2) In the IDP-VRMA architecture, the substrate network resource is allo-cated to many dynamic virtual network requests, affiliated with different SPs. These virtual network requests have different topologies, as well as the needs and constraints of various heterogeneous resource. Hence, there exists complex relationships among multiple VNs or between multiple VNs and InP. That is, multiple VNs simultaneously compete for different resource in different com-ponents of the substrate network. To model the complex relationship, this paper proposes a multidimensional non-cooperative game based virtual resource al-location mechanism. That using the utility function, price function and conges-tion function models the total pay function of each virtual network. Moreover, we prove the existence and uniqueness of the multidimensional Nash equilibri-um in the model. The simulation results show that the multidimensional Nash equilibrium can be achieved in various areas of the game model, the selfish be-haviors of the virtual networks are effectively suppressed, and virtual resource is fairly, reasonably, efficiently allocated.(3) The IDP-VRMA needs a stable and fast network traffic prediction al-gorithm with good prediction performance. For the problem, this paper pro-poses a deterministic echo state network with a loop-feedback reservoir, and applies it to network traffic prediction. The reservoir of the model is con-structed by introducing adjacent feedbacks between adjacent units based on the simplest loop reservoir structure. The absolute of the input weight is the same as the internal weights, and equal to r. Therefore, only one free param-eter is tuned, which greatly simplifies the echo state network. Furthermore, using state update equation and output equation, we prove the good memory capacity of the model. Simulation results show that the model can characterize the self-similarity of network traffic, and have good prediction performance.更多还原