【作者】 张冰怡；

【导师】 孙亚民；

【作者基本信息】 南京理工大学 ， 计算机应用技术， 2004， 博士

【摘要】 Internet网络的数学模型和QoS(quality of service)问题是NGI(next generation Internet)需要解决的两个课题。本文基于瞬时流速随机性假设提出了一个分数Alpha网络通信量模型，基于流和会话一级分别给出了证明。将模型和郎之万方程联系起来，给出了更多的物理解释。所提模型能体现网络通信量的自相似性和非高斯特征，并具有特定的物理意义。为了定量判断所提模型与真实通信量的拟合度，本文基于线性鉴别分析理论提出了一个判定算法，使用该算法分析了分数Alpha模型数据、真实通信量数据和NS仿真数据，验证了算法的有效性、优越性和普遍性。 基于分数Alpha模型，论文研究了QoS中的缓存计算、路由和接纳控制问题。推出了一个缓存溢出概率公式，该公式比基于其它网络模型得到的结果更符合真实数据的整体变化趋势。在高速路由器转发引擎的缓存设计中应用该公式得到了满意的结果，与传统方法比较，该方法更体现实际通信量变化规律，计算结果更准确。 基于多约束的QoS路由是一个NP完全问题，在算法复杂度上已有大量研究，但在路由策略和网络随机性方面考虑的还不多。同时随着全光网的发展，网络带宽瓶颈主要受路由器转发能力的限制，本文通过缓存溢出公式预测对转发能力的需求并用于QoS路由，提出了一个单播QoS路由算法，该算法能提高带宽利用率和网络吞吐量。另外，引入满意适应理论来理解和拓宽次优解概念，增加一个QoS参数变化概率初步考虑了网络变化的随机性。 已有接纳控制算法的拥塞概率基于传统电信网络的经验，假设服务时间符合指数分布，在数据分组网络中不一定合适。为此首先提出一个保证QoS的多流复用VoMPLS方案，以该方案为背景设计了一个具有可调系数的CAC算法，该算法通过带宽分配可调系数控制会话录用，保证高等级服务的需要。针对该算法，分别假设服务时间符合指数分布和考虑网络通信量变化的影响，基于分数Alpha模型进行了详细的分析，推导出了拥塞概率。结果表明两种方法得到的拥塞概率形式上是一致的，但基于传统方法的计算结果只是实际结果的下限。更多还原

【Abstract】 Internet mathematics model and QoS are two important issues in Next Generation Internet (NGI). In this paper, based on hypothesis for traffic rate a fractional Alpha traffic model is proposed. Two proofs based on flow level and session level is given separately. The model is also associated with Langevin equation. This traffic model can give the network character of self-simolar and non-Gaussain. For giving a judge of whether the traffic model fit the actual traffic, we import linear discriminant analysis (LDA) into traffic analysis and give a algorithm. Based on this algorithm we analyze data of fractional Alpha model, data of actual traffic, data of NS. Comparing with traditional method this algorithm is useful and can conveniently give an accurate judgment for complex network traffic trace.Based on fractional Alpha model, we research the computing of buffer size, QoS routing and call admission control (CAC). Firstly we deduced a formula for the residual of the queuing distribution function (RDF). This RDF can meet real RDF better than the RDF based on other models. Based on this formula, we compute the buffer size in high-performance routers. Our method is more accurate and simpler than traditional method and accord with actual network traffic.QoS routing with multi-constraints is NP-complete. There has been much research in reducing the computing complexity but there is lack of research in routing mechanism and routing randomicity. With the development of optic network, the bottleneck bandwidth will be limited by the performance of router forwarding. Based on the RDF formula, we predict the need for performance of forwarding and proposed a QoS routing algorithm. This algorithm can improve the utility of bandwidth and amount of traffic. We use satisfaction-adaptation scheme to expand the conception about sub-optimal paths. We add a parameter of probability to think of network randomicity.Traditional analysis method of blocking probability is based on the experience of telecom network and has a hypothesis of bandwidth holding time exponentially distributed which may be not appropriated in computer network. So firstly we proposed a VoMPLS traffic-multiplexing scheme with guaranteed QoS. Then we give CAC algorithm. This algorithm can guarantee high-grade service based on adjusting two coefficients. To this algorithm we deduce the blocking probability based on fractional Alpha model and traditional hypothesis separately. The expression of two blocking probability are similar but the result based on traditional hypothesis is only the lower bound of actual blocking probability.更多还原