Light-weight traffic parameter estimation for on-line bandwidth provisioning

Future network components will power down unused resources to save energy. Thereto, they need to determine the required capacity by observing the traffic. In this paper, we propose a light-weight estimator for the relevant parameters of aggregated packet traffic. The estimator assumes an M/G/∞ traffic model on flow level, which has been proposed for aggregated Internet traffic. We find that the variation of the aggregate traffic rate is defined by the bit rate of contributing application streams, i.e. traffic bursts triggered by end-user applications. We identify the effect of these streams on the variance-time behavior of the aggregated traffic rate. From this, we derive an estimator for the application stream bit rate based on second-order statistics of the aggregate rate. Simulation results for inelastic and TCP traffic show a good stream rate estimation accuracy, provided that the measuring period is sufficient to capture the variance of the aggregate rate.