New Solution Techniques for the Traffic Matrix Estimation Problem

We propose an approach which formulates the traffic matrix estimation problem as a non-negativity constrained optimization problem, and then a projection method is used to solve it. We conduct experiments both on synthetic and real measurement data obtained from the Abilene network. The results indicate that the estimated traffic matrices are more accurately estimated when our approach is used than when the tomogravity method is employed. We also develop a novel approach for estimating traffic matrices when optimal multi-path routing is employed. We formally formulate the problem as a bilevel programming problem. Then a genetic algorithm is used to solve it.