【作者】 吕博；

【导师】 刘韵洁；

【作者基本信息】 北京邮电大学 ， 通信与信息系统， 2011， 博士

【摘要】 互联网已深刻地影响着人们的生活和工作方式,对经济的发展、社会的进步做出了巨大贡献。但是,互联网及其体系结构经过40年的发展,其面临的可扩展性、安全性、移动性、服务质量、能耗等方面的一系列问题和困难逐渐凸显,难以适应全球网络规模的急剧扩张,难以满足层出不穷的新兴应用和新趋势的需求。网络虚拟化作为解决当前互联网僵化问题的技术手段,近年来受到了国内外未来网络领域研究的广泛关注。网络虚拟化的本质是通过抽象、分配、隔离机制在一个公共物理网络上独立地运营多个虚拟网络,从而能够有选择性地进行资源分配与调度,最大化物理网络资源利用率、提高服务质量、降低网络运营和维护成本。网络虚拟化在带来上述灵活性的同时,也给网络资源管理提出了巨大挑战。本论文分别从支持可扩展性、自治性的虚网资源管理架构,基于节点压力的虚网节点映射优化算法,以及基于链路压力的虚网链路映射优化算法等几个方面对虚网资源管理进行了研究,具体内容包括：1、针对网络虚拟化资源管理架构存在可扩展性不强、自治性差等问题,本文设计了一种层次型的虚网资源管理架构。该架构适用于底层多域物理网络环境,采用分层分域的管理机制,由全局调度中心执行接入控制、虚网划分、虚拟子网映射调度,各分域调度中心并行完成虚拟子网的资源映射及独立处理资源发现、故障管理、移动性管理等工作。这种集中式管理、分布式控制的结构设计,解决了集中式管理容易发生单点失效、性能瓶颈等问题,兼具了分布式系统的优势,能够很好地适应资源动态性、区域自治性等特点,大大降低了虚拟网络的管理复杂度,提高系统整体效率,保障虚网请求的实时处理。2、为了降低虚网映射算法问题求解的复杂度,大多数启发式算法将虚网映射分为节点映射和链路映射两个部分,并顺序求解。合理设计节点映射算法对于提升虚网整体映射性能具有重要影响。本文以节点压力为优化目标,提出了两种节点映射算法来改善虚网整网映射的性能。首先,受无线资源管理中的注水原理启发,提出一种最小节点压力优先的虚网节点映射算法。该算法通过对节点压力比较排序的方式对节点资源分配进行均衡控制,降低瓶颈节点出现的可能性,改善物理网络负载分布。仿真结果表明,该算法在获得相同映射效率的同时,可明显降低最大节点压力,均衡节点压力分布。其次,针对极限情况下的虚网映射问题,即虚网与物理网等规模的情况,提出一种最大节点压力优先的虚网节点映射算法。该算法通过最大化节点资源利用率,充分利用节点剩余资源,提高节点映射的成功率从而提升虚网整网映射的效率。仿真结果表明,该算法在虚网映射接受比和映射收益两方面均有明显提升,实现了物理网络资源的有效利用。3、基于链路压力状态反馈的思想提出了一种实现负载感知的自适应虚网映射算法。该算法是一种在线式的基于优先级区分调度的资源分配算法。该算法设计了物理链路剩余带宽反馈因子,有效区分了物理链路的映射接纳能力,通过将虚网映射函数与物理网络负载状态关联起来,实现了保障负载平衡的自适应资源分配。仿真结果表明,所提算法在取得相同映射效率情况下,能有效降低物理链路负载压力,将虚网请求均衡地映射在物理链路上,提高了资源利用率。最后,综合第四章提出的最小节点压力优先虚网节点映射算法与第五章提出的基于链路压力状态反馈的虚网链路映射算法进行联合试验,仿真结果表明两种算法的结合能够同时降低节点压力与链路压力,改善物理节点及链路的负载均衡,更为有效地提高物理网络资源利用率。更多还原

【Abstract】 The Internet has profoundly affected the way people live and work. It makes a great contribution on economic development and social progress. However, the architecture of Internet faces a range of issues and difficulties such as scalability, security, mobility, quality of service and energy consumption after 40 years of development. It is more difficult to adapt to the rapid expansion of the global network scale and meet the needs of new applications and traffic.Network virtualization technology plays an important role in solving the ossification problem of Internet. It supports diverse virtual networks sharing the common substrate network by abstraction and isolation mechanism essentially. Network virtualization allocates and schedules resource optimally to maximize the resource utilization and decrease the operation and maintenance cost. The above flexibility and advantage of network virtualization brings great challenge to the virtual resource management.This dissertation focuses on the scalable and autonomous resource management architecture, the efficient node mapping optimization algorithm based on node stress and the effective link mapping optimization algorithm based on link stress. The major works are outlined as follows:1. The resource management architecture of network virtualization has the issue of worse scalability and autonomy. To address these issues, we design a resource management architecture based on multi-domain and hierarchy. It adapts to the substrate network composed of multiple infrastructure provider domains. The architecture is composed of a Global Scheduler and several Local Schedulers. The Global Scheduler is responsible for the access control, virtual network partition and global embedding scheduling. And the Local Schedulers complete the sub-virtual network embedding in parallel and deal with the local virtual resource discovery, failure and mobility management independently. The hierarchical architecture reduces the management complexity, improves the efficiency and satisfies the real-time requirement of virtual network embedding. Additionally, it takes advantages of both the central and distributed resource management architectures. It can not only solve the single point failure and performance bottleneck problems caused by centralized models but also reflect the dynamic and autonomous characteristics of resource very well.2. Existing virtual network embedding algorithms are usually decomposed into two steps:virtual node mapping and virtual link mapping. This decomposition helps reduce the overall complexity of resource allocation for virtual network requests since the virtual network embedding problem is NP-hard. Embedding the virtual node efficiently can promote the performance of the whole virtual network embedding algorithm greatly. We propose two different node mapping algorithm based on node stress to promote the performance of virtual network embedding. First, a node mapping algorithm based on minimum node stress priority is proposed to improve the load balancing of substrate node. It sorts of node stress in order to allocate the node resource balanced. Simulation experiments show that the proposed algorithm achieves the same embedding efficiency and improves the load distribution of substrate nodes distinctly while reducing the average substrate node stress significantly. Another novel node mapping algorithm is proposed based on the maximum node stress priority when the scale of virtual network is equal to that of substrate network. It aims to maximize the substrate node utilization in the node mapping stage to improve the efficiency of virtual network embedding. Simulation experiments show that the proposed algorithm improves the revenue and acceptance ratio of virtual network embedding significantly.3. A load-aware virtual network mapping algorithm is proposed based on the status feedback of link stress. It deals with the online virtual network requests with priority scheduling defined by mapping revenue. The main contribution is to embed the virtual network requests according to the current load distribution of substrate network. This adaptive algorithm differentiates the residual bandwidth of substrate links and takes full advantage of the multi-path to improve the load balancing of the substrate network by the status feedback factor. Simulation experiments show that the proposed algorithm achieves the same embedding efficiency and improves the load distribution of substrate network distinctly while reducing the average substrate link stress significantly. Finally, we combine the node mapping algorithm based on minimum node stress priority and the link mapping algorithm based on status feedback of link stress to do simulation. The results show that the combined algorithm can reduce the maximum node stress and maximum link stress simultaneously and improve the node load balancing and link load balancing significantly.更多还原