Research Of Traffic Matrices Estimation Methods Based On IP Backbone Networks

With the rapid development of Internet, the scale of the network has become more and more large, and the structure has become more and more complex, and the number of Internet users has grown exponentially. However, the non-critical business in the network has also largely consumed the bandwidth resources of network, affecting the running of other critical business. These have led to its monitoring and management increasingly difficult. In order to carry out better network planning, network design, network management, network monitoring, routing configuration, network traffic engineering and network simulation, the traffic information of network is needed urgently. Traffic matrix, which is an overview of the whole network, is a complete description of traffic flows and its distribution. Combined with network routing information, it can also clearly reflect traffic component of each link in the network. It is a key input parameter of network traffic engineering and network management. However, large-scale, traffic matrix is difficult to obtain through direct measurement method in the complex network. Currently, estimating traffic matrix through the limited measured information has become the main method. The problem of traffic matrix estimation is an ill-posed linear inverse problem.This article describes the development process of traffic matrix estimation. Representative of every stages of traffic matrix estimation methods are described in detail. We analyze the advantages and disadvantages of each method. The innovative achievements of this article are in the following two aspects.For ill-posed characteristics of traffic matrix estimation, we avoid the traditional idea of estimation algorithm, namely, increasing the constraint condition by modeling the origin-destination flows to overcome the constraints of ill-posed characteristics. Through statistical analysis in large quantities of the actual measured traffic matrix data, we assumpe that traffic matrix has the characteristic of spatial self-similarity. We proposed linear mapping method based on spatial similarity. Experiments show that the algorithm calculates quickly and accurately.Complex nature of network traffic makes the current researchers tend to use more sophisticated and complex model for traffic matrix estimation, neural network is a good way. On account of the traditional neural network’s trouble of memory lost or distorted in traffic matrix estimation, we posed a traffic matrix estimation method based on multi-neural network. This method improves the training efficiency and estimation accuracy.