A Novel Network Tomography Approach for Traffic Matrix Estimation Problem in Large-Scale IP Backbone Networks

A traffic matrix is an important input for some network management function such as traffic engineering and network planning. However, traffic matrix usually cannot be obtained directly. That is because the existing network traffic motoring techniques do not extremely well in practice. The common approaches for obtaining a traffic matrix are estimating it via other available information, such as link loads and routing information. The approach using link loads and routing information to infer the traffic matrix is named network tomography approach. Due to the ill-posed nature of the network tomography model, the network tomography techniques still face many challenges for the traffic matrix estimation problem. Motivated by this issue, we propose an optimization model to conquer the ill-posed nature of the network tomography model. In details, we first perturb the network tomography model by a stochastic matrix, which draws on idea from compressive sensing techniques. Then the perturbed model can be solve by a convex optimization model. However, this convex optimization cannot precisely estimate the traffic matrix since the coherence of the perturbed model. Then we proposed an improved optimization model by means of the network tomography model.