On-line Measurement of QoS for Call Admission Control

On-line measurement of tra c can improve the quality of network monitoring and call admission control algorithms by providing reliable estimates of QoS parameters. We propose to estimate QoS by tting an appropriate function to the bu er occupancy distribution of a singleserver rst-comerst-served queue. We collect samples of the bu er occupancy distribution just after packet arrivals in the queue. Then, we t an approximating function, a positive sum of exponentials, using minimum chi-squared estimation techniques. The QoS parameters can then be inferred from the tted distribution function. We consider three types of minimum chi-squared estimators: the rst uses the sample covariance matrix, while the last two use biased approximations that are considerably easier to compute. Furthermore, we extend these algorithms to virtual queuing systems to estimate the impact of additional connections being admitted to the queue. We provide algorithms and criteria for all three estimators, along with simulation results. These simulations show that we can obtain reliable estimates of QoS parameters within minutes.