Research On Markov-BPNN Estimation Model And Method Of Traffic Matrix

With the rapid development of information technology,4G, WiFi and WiMax technologies have been successfully applied to network. The number of IP network and mobile Internet terminal users is growing exponentially, which make the current network more and more complex. In addition, network traffic shows diversity, a lot of new features and heterogeneity that make it more difficult to manage. Traffic matrix is an important input parameter for operators to manage network, routing optimization, congestion control, network detection, network configuration, load balancing, traffic detection and fault diagnosis. However, it is very difficult and even impossible to measure the traffic matrix form the large scale of network directly. The relationship between the link measurement matrix, the routing matrix and the traffic matrix can be expressed by a constraint equation. However, in the large scale networks, since the number of OD(Origin-Destination) pairs is much more larger than the number of IP network links, thus the constraint equations showed less qualitative with numerous solutions, which leads to a high degree of ill posed feature in the traffic matrix estimation problem. The main challenges of the traffic matrix estimation problem are how to overcome this ill posed feature and accurately estimate the traffic matrix.For the traffic matrix estimation problem, we need not only consider the characteristics of the traffic matrix constraint equations, but also take into account the characteristics of the traffic matrix itself. As the research further developed, researchers found out that traffic matrix has a variety of complex and changeable characteristics, such as temporal correlation, spatial correlation, heavy-tailed distribution, self-similar, short and long-related and so on. To solve the traffic matrix estimation problem, a new method named Markov-BPNN estimation method was presented in this paper. Firstly, to overcome the ill condition of constrained equation and take into account the temporal correlation of the traffic matrix, in this paper, the constraint equation of the traffic matrix was transformed into the equation of the link measurement matrix which is constrained by the random matrix and the routing matrix. By studying the Markov process of the random matrix, the Markov estimation method of the traffic matrix is obtained. This method not only avoided the ill posed feature of constraint equations but also captured temporal correlation characteristic of the traffic matrix accurately. Secondly, based on the estimated traffic matrix by Markov processes of the random matrix, a multi layer feed forward neural network which is trained by the most widely used back propagation (BP) algorithm is applied to improve the accuracy of the estimator. The Markov-BPNN estimation method of the traffic matrix makes the estimated values closer to the real values and improve the accuracy of the estimation effectively. The results of simulation show that, Markov-BPNN method has better performance than generalTomogravity which is considered as a well-known accurate method. Finally, the simulation results also demonstrate the advantages of Markov-BPNN method model in practicability, accuracy and robustness even if there exists random interference link traffic matrix.