Research And Simulation Of Optimization Algorithm For Traffic Matrix Estimation In Backbone Networks

With the rapid development of information technology and IP network, the size of network rises by exponential form. IP network has become heterogeneous, open and complex. Network traffic presents some new characteristics, for example, spatio-temporal correlations, self-similarity, long and short range dependence and heavy-tailed distribution. As the complexity and heterogeneity of IP network, it’s difficult to obtain the traffic matrix of IP network by the direct measurement and even is not practical sometimes. So according the link load to estimate the traffic matrix becomes one hotspot of the current researches.Traffic matrix estimation holds highly ill-posed nature. In order to overcome the ill-posed problem, a traffic matrix estimation method based on genetic algorithm is proposed to overcome the ill-posed nature and spatio-temporal correlations nature problems. The algorithm exploits global self-adpating search to solve the problems existing in traffic matrix estimation. It exploits the constraints of IP traffic matrix and the characteristics of genetic algorithm to obtain the global optimal solution. It enhances the speed of convergence and computing in the optimal process. And it takes the real network datas as the initial value to obtain the accurate estimation of the large-scale IP traffic matrix.Traffic matrix estimation based on principal component analysis and decomposable principal component analysis algorithms are proposed to solve the high ill-posed nature and spatio-temporal correlations nature problems. It exploits principal component analysis to reduce the dimension and the scope of the solution dimension. And the high ill-posed problem can be translated into posed problem and reduce its ill-posed nuture. Decomposable principal component analysis algorithm exploits the the prior information in these models to distribute its computation. And it reformulates the problem in the sparse inverse covariance domain and solves the global eigenvalue problem using a sequence of local eigenvalue problems in each of the cliques of the decomposable graph. Finally, we exploit traffic data from Abilene network to validate our method. Simulation results show that our approach can accurately estimate large-scale IP network traffic matrix.A mixed algorithm of genetic algorighm and principal component analysis algorithm are proposed to solve sensitiveness to the initial value of genetic algorithm and the bad adaptivity of principal principal analysis algorithm. It composes two parts:firstly, it exploits principal component analysis to obtain the principal components; secondly, it takes the princiapal components as the initial value of the genetic algorithm. Simulation results show that our approach can accurately estimate large-scale IP network traffic matrix.