网络流量的新模型研究

发布时间：2018-06-02 22:47

本文选题：流量矩阵 + 压缩感知　；参考：《北京交通大学》2014年硕士论文

【摘要】：互联网技术作为21世纪发展最快的技术之一,已经广泛应用到人们的生产、生活当中,对社会的进步、经济的发展做出了巨大的贡献。然而,随着互联网技术的进一步成熟,近年来也涌现出了大量的新型网络应用和服务,它们在给人们带来便捷的同时,也给网络运营商的维护管理带来了巨大的压力。与此同时,数量众多的异构网络的接入,使得互联网变得更加难以掌控。通过网络测量能够获知网络性能的一些重要参数,从而可以更好的管理网络。因此,如何有效的进行网络测量成为了一个重要的研究课题。流量矩阵是网络测量的一个重要参数,它反映了一个网络中所有源节点和目的节点对之间的流量需求,是网络规划和流量工程的一个重要输入。目前常利用网络层析成像技术估算流量矩阵,它是从链路级的测量数据中估算出路径级的参数。网络层析成像技术的本质决定了流量矩阵估算问题是一个欠约束问题,这就需要增加一些先验信息作为约束条件以得到最优解,由此衍生出了许多先验模型。由于多数大型网络的链路测量值有限,以往的先验模型不能依据少量的链路信息有效的估算出流量矩阵,因此不能应用于大型网络。压缩感知理论表明任何充分可压缩信号能够利用少量非适应性的随机线性投影样本进行重构。该理论适用于流量矩阵的估算,于是提出了一种基于压缩感知技术的概率模型。概率模型能够从少量的链路信息中重构出流量矩阵,因此适用于大型网络的流量矩阵估算。本文通过仿真实验验证了概率模型的有效性,并使用真实网络的流量数据,对比概率模型和经典的重力模型,证明了概率模型的估算效果更好。将概率模型应用于真实网络的流量矩阵估算时,难点在于如何确定符合真实网络的概率参数。为了解决这一问题,本文提出了重力-概率模型。利用重力模型估算得到的流量矩阵计算概率参数,然后将该参数用于概率模型。使用真实网络数据的仿真实验表明,重力-概率模型的估算效果优于普通的概率模型和重力模型。
[Abstract]:As one of the fastest developing technologies in the 21st century, Internet technology has been widely used in people's production and life, and has made a great contribution to social progress and economic development. However, with the further maturity of Internet technology, a large number of new network applications and services have emerged in recent years, which not only bring convenience to people, but also bring great pressure to the maintenance and management of network operators. At the same time, access to a large number of heterogeneous networks makes the Internet more difficult to control. Some important parameters of network performance can be obtained by network measurement, which can better manage the network. Therefore, how to effectively carry out network measurement has become an important research topic. Traffic matrix is an important parameter of network measurement, it reflects the traffic requirement between all source nodes and destination nodes in a network, and it is an important input of network planning and traffic engineering. At present, network tomography is often used to estimate the flow matrix, which estimates the parameters of the path level from the measurement data at the link level. The nature of network tomography technology determines that the flow matrix estimation problem is an underconstrained problem, which requires the addition of some prior information as a constraint condition to obtain the optimal solution, and many prior models are derived. Due to the limited value of link measurement in most large networks, previous prior models can not effectively estimate the flow matrix based on a small amount of link information, so it can not be applied to large networks. Compression sensing theory shows that any fully compressible signal can be reconstructed using a small number of random linear projection samples. This theory is applicable to the estimation of traffic matrix, so a probability model based on compressed sensing technology is proposed. The probability model can reconstruct the traffic matrix from a small amount of link information, so it can be used to estimate the traffic matrix of large networks. In this paper, the validity of the probabilistic model is verified by the simulation experiment, and the probability model is compared with the classical gravity model by using the flow data of the real network, and it is proved that the probabilistic model is more effective. When the probabilistic model is applied to the estimation of the real network flow matrix, the difficulty lies in how to determine the probability parameters in accordance with the real network. In order to solve this problem, a gravity-probability model is proposed in this paper. The probability parameter is calculated by using the flow matrix estimated by gravity model and then applied to the probability model. The simulation results using real network data show that the gravimetric probability model is better than the ordinary probability model and gravity model.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06

【参考文献】