Network Planning: Traffic Matrices Estimation and Demand Uncertainty

The traffic matrix is the fundamental input data in network pl anning, simulation and traffic engineering. However, it is often unknown and its direct measurement with devices such as Netflow is a too heavy process for large high-speed net works. The estimation of the traffic matrix appears as the best alternative approach. How ever, planning a network using a single “busy hour” estimated traffic matrix strains credib il ty because with the increasing popularity of higher bandwidth applications, traffic patte rns are more and more volatile even in the aggregate. Such an approach can lead to a poor utilizat on of network resources if at some point in time the actual traffic matrix deviates signific antly from the one used for route optimization. In this paper, we address the problem of netwo rk planning under uncertainty on the estimated traffic matrices. We introduce a novel demand u ncertainty model. Our model provides a concrete uncertainty structure that makes sense for a network operator. We study then two network planning problem under our demand uncertai nty model: (1) the link weight optimization problem in the OSPF and IS-IS networks, and (2) LSP routing problem in the MPLS networks.