Packet loss probability estimation with a Kalman filter approach

Provisioning QoS enabled MPLS VPN services in IP networks has attracted a lot of attention due to the high Return on Investment (ROI). In spite of considerable research effort, no practical solution to this problem has been found One of the main issues to be solved is to estimate the Packet Loss Probability (PLP) accurately and effectively based on the input stochastic traffic process. Inspired by the Large Deviation Theory (LDT), two types of asymptotes loss estimators have been studied in the practical MPLS VPN networks: the Large Buffer Estimator (LBE) and the Aggregate Traffic Estimator (ATE). In both of the estimators, the traffic mean and variance have to be estimated as much as accurately possible. Clearly, tracking of the traffic mean and variance is central in the estimators. In this paper a Kalman filter is applied to optimally recursive estimate the traffic mean and variance. Kalman filter is a general method for the optimal estimation of a noisy measurement, using the estimation error obtained from the past measurement to fix the one-step prediction. The algorithm runs recursively and is applicable for the on-line application. A series of experiments evaluate its performance oil the live NCIT*net2 network under different traffic arrival models and different buffer sizes. The numeric results verify the effectiveness of the algorithm.