Combining sFlow and Tracker traffic analysis: A novel estimation approach for network-wide BitTorrent distribution

BitTorrent(BT) has emerged as one of the most popular protocols for content sharing in recent years. Most BT applications are network-oblivious which brings great challenges to traffic engineering. As basic input information, network-wide BT traffic distribution is of vital importance for service providers (SP) or carriers to enforce appropriate control policies. Direct online measurement of network-wide BT traffic faces scalability issue, for it requires deploying application identification deep packet inspection (DPI) device on every subject link. In this paper, we propose a novel approach to estimate network-wide BT traffic distribution through combining sFlow and Tracker traffic analysis. Owing to a noticeable increase of BT traffic using random port numbers, sFlow alone is no longer sufficient for BT recognition. We tackle this problem by combining sFlow with information collected from BT tracker´s traffic. With assistance of this information, BT samples can be accurately distinguished from other sFlow samples. The efficacy of our approach is evaluated through experiments on backbone links of campus network. Evaluation results show that our approach can reach a high accuracy in network-wide BT traffic estimation.