A stochastic programming method for IP traffic matrix estimation

Traffic Matrices, which capture the network traffic volumes among end-points, are widely used in Internet Engineering. A traffic matrix is generally estimated from link loads because it is impossible to measure each flow directly in a large scale IP network. The estimation problem is usually considered to be an optimization model with a routing constraint which relates the traffic matrix with link loads. However, the information provided by link loads is far less than that required, and so the estimation problem becomes very ill-posed. Moreover, the random measuring noise of link loads often makes the route constraint violated. To deal with this problem, this paper proposes a stochastic programming (SP) model, which relaxes the routing constraint to a probabilistic constraint and then enlarges the feasible space of the optimization problem. By using the Abilene dataset, we simulated the proposed SP model and the Tomogravity method respectively. The simulation results show that the SP model can improve the estimation performance obviously.