Research On IP Traffic Matrix Estimation By Stochastic Programming Method

The Internet technology is now one of the fastest growing and the most widely used technology, but in recent years, with the development of the Internet technology gradually mature, the further popularization of Internet application, and the application of new network services in the Internet such as bamboo shoots after a spring rain like emerge. The huge network scale, a large number of data transmission link, access to many heterogeneous networks, allow network researchers to directly measure to obtain the network traffic matrix difficultly.Traffic matrix is an important support for many of the network technology,the network path reflects the each flow of demand, have important applications in many engineering fields. We must find a new way to solve problems and effectively, fastly access to the network traffic matrix, lay a good foundation for the next research network.The paper introduces the concept and method to obtain the traffic matrix. The problem of traffic matrix estimation method and related models are summarized, mainly introduces the tomography technology and gravity model. By directly measuring the traffic matrix approach is not feasible, only through the estimation of the way to get, the main task of this article is to solve the problem of traffic matrix estimation. In the traffic matrix estimation equation, the number of OD flow is far greater than the number of links in the IP network, so that this equation is underdetermined, ill posed equation, so difficult to solve. And all the before models are in the ideal situation,without the existence of link noise.This paper presents a new model, a stochastic programming model. By the introduction of a random variable of constraint function in traffic matrix estimation equation,the equation is turned into the probability equation, in order to increasing the solution space of the equation,thus increasing the possibility of finding the optimal solution, and is the key link, the random variables representing the network noise, can better simulate the real network environment. Through comprehensive theoretical analysis and based on the real network data, and with the classic model of gravity tomography simulation experiment results, we can come to conclusion that the estimating effect of stochastic programming model is better, and more close to the actual value of the network.