Studying the Effective Bandwidth through the Distribution of One-Way Delays

This paper describes a method of estimating the effective bandwidth through modeling the tail probability of one-way cumulative variable delays. In order to estimate the effective bandwidth of a fixed one-way path, a sequence of probing packets are sent from a source to a sink along the one-way path. The one-way cumulative variable delays are first estimated based on the local time measurements made at the two end nodes with respect to their skew-free local clocks. The modeling of the distribution of one-way cumulative variable delays is based on the assumption that the traffic pattern along the one-way path follows a Gaussian distribution. Under the assumption of Gaussian traffic, the tail probability of the one-way cumulative variable delays can be expressed in an exponential form. The effective bandwidth of a fixed one-way path is derived from the exponent of the exponential-form tail probability of the one-way cumulative variable delays. Our method has been numerically analyzed using data samples collected from network simulations. The numerical analysis shows that, at lower delay bounds, the effective bandwidth estimated from our modeling is quite comparable to the effective bandwidth calculated from the empirical tail probability of the one-way variable delays measured in simulations. At the higher delay bounds, the effective bandwidth estimated from our modeling is projected to be more accurate than the the effective bandwidth calculated from the empirical measurements.