【作者】 漆华妹；

【导师】 陈志刚；

【作者基本信息】 中南大学 ， 计算机应用技术， 2010， 博士

【摘要】 无线Mesh网络(Wireless Mesh Networks, WMNs)是一种高容量高速率的分布式网络,不同于传统的无线网络,可以看成是一种WLAN和Ad Hoc网络的融合,且发挥了两者的优势。具有可靠性、自组织性和自愈性等特点。随着人们对网络通信需求的增加,人们对无线Mesh网络的要求越来越高,并且希望能够提供QoS保障。因此,无线Mesh网络的QoS性能研究显得越来越迫切和重要。然而,无线Mesh网络的无线多跳通信以及带宽受限等特点,都使得在无线Mesh网络中QoS的研究困难,给广大研究人员提出了挑战。研究提高无线Mesh网络性能,有效实施网络QoS控制,避免网络拥塞,具有重要的研究价值。本文以无线Mesh网络的QoS性能问题为主要研究对象,在对无线Mesh网络的流量特征进行深入分析的基础上,针对无线Mesh网络中无线链路不稳定、信道误码率高以及流量的统计性等特点,利用网络演算(Network Calculus)理论系统地研究了基于网络演算的无线Mesh网络QoS性能模型。本文的主要研究成果和创新点如下：(1)基于随机／统计网络演算理论,构建了无线Mesh网络流量模型WgSBB(Wireless generalized Stochastically Bounded Burstiness)。在总结分析目前国内外网络流量模型主要研究成果的基础上,基于随机／统计网络演算理论,提出了一种适合于无线Mesh网络的业务流的流量模型WgSBB,该模型采用队列长度进行统计描述,构造了无线Mesh网络单节点和多节点的WgSBB流量模型,其突发函数受限于一个最小加卷积函数,表明其具有较严密的边界函数；阐明了它的求和特性和输出流量特征及其输入输出关系,从理论上证明了每个资源的流量特征,研究表明该模型能正确地描述无线Mesh网络业务流量特征的实际情况。(2)基于随机／统计网络演算理论,提出了适应于无线Mesh网络网关通信量的贪婪分形漏桶整形器(Greedy Fractal Leaky Bucket, GFLB),推导了GFLB的性能。基于随机／统计网络演算理论,给出了适应于无线Mesh网络网关通信量控制的贪婪分形漏桶整形器GFLB的一般数学模型,并推导了GFLB的输出特性、队列长度与延迟等性能,讨论了GFLB的引入对网络端到端延迟以及数据丢失总量的影响,并得出GFLB的引入并不增加系统端到端延迟上界、不增加缓存容量需求等结果。这些分析工作和所得结果对于基于GFLB的业务流控制方案的评价和如何有效地调节和控制网关节点通信流具有实际意义。(3)基于随机／统计网络演算理论,推导了基于GFLB的无线Mesh网络网关节点QoS性能随机/统计上界模型。网关是无线Mesh网络的性能瓶颈,网关节点的缓存区队列长度上界、延迟上界以及延迟抖动上界在网络规划、设计及配置中具有重要意义。基于已有研究成果,引入GFLB对传输到网关的网络业务流进行整形,运用公平的带宽分配策略,分析了网关节点的QoS性能随机/统计上界模型,给出缓存区队列长度随机／统计上界、延迟随机／统计上界和延迟抖动随机／统计上界。数值分析结果表明：基于GFLB的网关可用带宽能公平地分配给各会话流,并隔离不同的会话流,各会话流互不影响；基于GFLB对网关流量整形,结合公平的带宽分配策略,能为无线Mesh网络网关业务流QoS提供性能上界保证；利用网络演算理论计算基于GFLB的网关QoS性能边界问题具有简单性和实用性。(4)基于随机／统计网络演算理论,推导了网关与Mesh路由器端到端QoS随机／统计上界模型。网关与Mesh路由器之间端到端QoS性能,其上界能否得到保证将直接影响拥塞和接纳等网络控制算法与策略的有效性。在对网络端到端QoS性能研究方面,利用随机／统计网络演算理论计算网关到Mesh路由器之间端到端通信流QoS性能上界,推导出基于缓存队列共享的无线Mesh网络单节点的队列长度随机／统计上界以及端到端队列长度随机／统计上界、单节点的延迟随机／统计上界、端到端延迟理想随机/统计上界、端到端延迟近似随机／统计上界、端到端延迟抖动随机／统计上界。证明了基于端到端有效服务曲线的无线Mesh网络通信量端到端延迟随机／统计上界具有更好的紧致性。数值分析结果分析表明,基于有效服务曲线和网络演算理论能很好地对无线Mesh网络通信流端到端随机／统计上界进行计算。这对无线Mesh网络环境中提供随机/统计的服务保证的有效控制、调度和管理具有参考价值。更多还原

【Abstract】 Wireless Mesh Networks (WMNs) is a high capacity and transmission rates distributed network organized in a mesh topology. The difference from conventional wireless networks are WMNs combine WLAN and Ad Hoc networks and possess the advantages of both with reliability, self-organizing and self-healing characteristics. Along with the increasing demand of network communications, more and more reliable WMNs are required. WMNs Quality of Service (QoS) enhancement research, therefore, is important. However, the characteristics of multi-hop routing and bandwidth fairness bring many challenging problems in QoS research. In order to enhance the quality of networks, the research on how to efficiently implement QoS control for WMNs and how to prevent network congestion collapse, has profound impacts.In this dissertation, we focus on the QoS performance concerns in WMNs, and carry out our in-depth study the characteristics of traffic in WMNs. Aimming at the WMNs characteristics of instability wireless link, high probability wrong code channel and statistical traffic, this thesis employs Network Calculus theory to elaborate a WMNs QoS performance evaluation model. The research contributes to the development of the body of knowledge surrounding WMNs QoS. Its contributions include the following:A new proposed WMNs traffic model WgSBB based on stochastic/statistical network calculus.Having reviewed the literature on network traffic modeling, a new traffic model WgSBB is proposed for WMNs based on stochastic/statistical network calculus. It adopts backlog bound as descriptive statistics to represent a single and multiple node(s) traffic model for WMNs. The busty function of the model is constrained under a min-plus algebra, which indicates the model has a rigorous boundary function, then it illustrates the possession of summation and in/outbound traffic characteristics, and the relationship of inbound and outbound traffic. It has been theoretically proved that the model is able to properly depict WMNs traffic characteristics.A new proposed greedy fractal leaky bucket shaper GFLB and its performance parameters for WMNs gateway traffic based on stochastic/statistical network calculus.In this thesis, a mathematical model of the greedy fractal leaky bucket shaper(GFLB) is proposed for WMNs to optimize or guarantee performance, and a number of corollaries of the greedy fractal leaky bucket shaper are derived, such as input/output, queue length and latency characteristics. Also, a thorough arid sound discussion of the impact of integrating the greedy fractal leaky bucket shaper into network edges on communication latency and the total packets losing is conducted, and it concludes that the migrated greedy fractal leaky bucket shaper has no negative impact on the WMNs. The discussion significantly contributes to other researchers on how to evaluate the greedy fractal leaky bucket shaper traffic control solution and how to effectively adjust and control Mesh gateway traffic.A new proposed QoS performance stochastic/statistical upper bound model for GFLB WMNs gateway nodes.In WMNs, Mesh gateways are the performance drawback. Upper bound on cache backlog bound, delay and jitter of Mesh gateway node, therefore, play important roles in WMNs planning, designing and deploying. According to the art of the state in WMNs, in this thesis, greedy fractal leaky bucket shaper and strategy for bandwidth distribute fairly are applied at the network edges to analyze the QoS performance statistics upper bound model to assess the upper bound of cache backlog bound, delay and jitter. The experimental data indicates that applied greedy fractal leaky bucket shaper in WMNs gateways can fairly allocate network bandwidth to all network sessions without interfering. With the proposed model, the WMNs traffic QoS performance upper bound can be guaranteed, and it is simple and applicable.A new proposed end-to-end QoS stochastic/statistical upper bound model for WMNs gateways and Mesh routers.The end-to-end QoS performances between Mesh routers and Mesh gateways are determined in the mean of whether upper bound can be guaranteed. It can directly affect the network control algorithms of network congestion collapse. Therefore, in this thesis, the employment of network calculus to calculate the Mesh routers end-to-end QoS performances upper bound can conclude the stochastic/statistical upper bounds of WMNs single node backlog bound, delay, idealized delay, approximate delay and jitter. The conclusion proves WMNs traffic delay statistics upper bound on service curve has better compactness. The experimental data indicates that the effective service curve and statistics (stochastic) network calculus are suitable for WMNs, and it contributes the development of WMNs traffic control, plan and manage for stochastic/statistical services.更多还原