Internet traffic engineering for partially uncertain demands

One of the weaknesses in conventional routing management is that it bases on given and often "worst case" traffic matrices. Imprecision in a single element of the traffic matrix could result in inefficient routing decisions and unpredictable performance forecast errors. As the number of communication endpoints and diverse applications grows increasingly, the task to accurately forecast demands is becoming more and more difficult. Therefore, explicitly including traffic variations when making long-term network planning decisions as well as medium-term network provisioning policies has recently attracted much attention. In this paper, we consider an offline traffic engineering (TE) problem in IP networks for partially uncertain demands to address a situation where traffic is composed of both fixed and uncertain parts. This model is particularly appropriate for dealing with two different types of demands simultaneously i.e. (i) the fixed part of demands that have to be guaranteed according to some service level agreements; and (ii) the uncertain part of demands that vary over time. The proposed model improves efficiency compared to the case where all traffic is considered as uncertain, while still allowing traffic variations. We present several computational results for both multiple and unique shortest path routing, which show the benefits of our model.