Experiment and analysis of active measurement for packet delay dynamics

Active measurements have formed the basis for much of our empirical efforts to understand Internet packet delay dynamics. Packet-average performance of user flow is getting more and more important for users, and especially for network service providers. But in network active measurement area, there are not empirical efforts to investigate the performance of active measurement for packet performance from user's standpoint. We quantitatively assess and compare the one-way delay statistics experienced by user flow and active probe flow based on simulation experiments, and find that: (1) Active measurement systematically underestimates statistics of packet delay experienced by user flow and the estimation error is far severe than can be ignored. (2) Increasing sampling frequency is almost helpless for the reducing of the estimation error. (3) The estimation error degrees of active measurement decrease as the increasing of the queue utilization. The above conclusions are based on active measurements using Poisson sampling and Periodic sampling. As they are mainly used sampling methods in active measurement area, so our conclusions indicate that current active method for measuring Internet packet delay suffer from system errors from user's standpoint.