Online Packet Loss Measurement and Estimation for VPN-Based Services

When provisioning quantitative quality of service (QoS) for virtual private network (VPN) services over packet switched networks, parameters such as packet loss, delay, and delay jitter, besides the required bandwidth, have to be guaranteed. While the bandwidth is relatively easier to guarantee, maintaining a value of the packet loss parameter below a preset value presents great interests but serious difficulties. One of the key issues is to link the stochastic characteristics of the input process to the packet loss probability (PLP), i.e., how one can accurately estimate the PLP based on measurements of the input process. This is crucial for building transducers for control loops meant for keeping the packet loss parameter within the guaranteed limits as specified by the service level agreement (SLA). Although the estimation of the PLP has been studied by many researchers, little has been done in regard to the estimators of the packet loss parameter such that the latter can be used in online applications. This paper studies the PLP estimation problem from a transducer and, hence, a control system perspective, and evaluates the quality of the proposed solutions through live network experimental data. Two asymptotic estimation formulas for loss probability are derived by applying the large deviation theory (LDT) on the buffer overflow probability and proposed as mathematical relations to be used by the online transducer.