QoS-aware bandwidth provisioning for IP network links

Bandwidth provisioning is generally envisaged as a viable way to support QoS in IP networks. To guarantee at the same time cost-efficient use of resources, the crucial question is: what is the minimal bandwidth provisioning required to ensure the QoS level agreed upon (for instance: the probability that the traffic supply exceeds the available bandwidth, over some predefined interval T, is below some small fixed number epsilon. This paper deals with this dimensioning problem, with as crucial novelty that the resulting guidelines are based on coarse traffic measurements. Our approach relies on a powerful `interpolation' formula that predicts QoS on relatively short time scales (say the order of 1 s), by just using large time-scale measurements (say in the order of 5 m, as in the case of the standard MRTG measurements). As a result, we find that, measuring a load rho (in Mbit/s), the required bandwidth (to meet the QoS criterion) has the form rho +alpha sqrt{rho}, where alpha depends on T and epsilon -- this expression is derived under minimal model assumptions. Apart from its simplicity, the dimensioning formula has a number of attractive features, viz. its insensitivity and robustness (as just the load rho is needed), and its transparency (the impact of changing the `QoS-parameters', i.e., T and epsilon, on alpha is explicitly given). The dimensioning rule is validated through extensive measurements obtained in several operational network environments.