Research On Traffic Matrix Estimation Based On CRO High Order Neural Network

With the rapid development of Internet technology, more and more users join the Internet, so that the scale of network continues to expand. A variety of network applications emerge in an endless stream, with network traffic transmission growing in geometric. For Internet Service Provider(ISP), network operators and network management, cost saving, optimizing network planning and design, improving the quality of network service, to fully know the running state of the network well and other network engineering problems must be solved further. Understanding the various characteristics of the internal network contributes to design, to control and to manage network successfully. Network traffic matrix as one of the important tools, describes the network traffic distribution among all nodes comprehensively. It is an important basis for network design, management and routing configuration. However, the huge network scale, transmission data, heterogeneous distributed network model make traffic matrix through network measurement directly more and more difficult to get so that it is impossible. So many scholars put forward the indirect measurement method of network traffic matrix estimation by using limited measurement information.The main work of this paper focuses on traffic matrix estimation is as follows. 1)First, it introduces the research significance of traffic matrix, domestic and foreign research status, outlines the hierarchical structure of this paper.2)Second, it summarizes three types of methods for indirect traffic matrix measurement, and having a basic description of the technology applied in each method, analyzing their merits and drawbacks.3)Third, it applies the Chemical Reaction Optimization to high order neural network algorithm, puts forward a novel CRO-PSNN algorithm to estimate the traffic matrix. In this paper, it discusses the algorithm’s superiority from two aspects. For one thing, the natural advantages of high order neural network for traffic matrix estimation which with high dimension and is ill-posed. With the multiplier, high order neural network can deal with higher order problems and nonlinear problems which ordinary neural network can not. For another, applying the optimization algorithm in the learning process of high order neural network, avoids the error take part in the weight adjustment, reduces the amount of calculation, accelerates the convergence speed and computational speed of the neural network.4)At the end, according to the existing theoretical and experimental basis, it proves the advantage of this method by the contrast experiment with a well-known traffic matrix estimation method. In the final summary, we review the research work which has been done in this paper, and prospect further research according to the present situation.