Measurement and enhancement of BitTorrent-based video file swarming

Abstract

BitTorrent, the most popular P2P file sharing application over the Internet, has also been widely used for video file distribution, albeit in a download-and-play mode. In this paper, we extensively examine the characteristics of BT swarms with different contents through a large-scale measurement. Our study especially focuses on the existing video file swarms, trying to understand the potentials and challenges of providing streaming service in BitTorrent. Our results from intra-swarm, inter-swarm and AS-level measurements demonstrate that the video file swarms in the BitTorrent system are quite different from the non-video swarms. In particular, the preference bias terminates the long-term relationship between peers and raises a significant challenge to develop BT based streaming service. Fortunately, we find that most peer sets in video file swarms are good enough to support streaming service. Strong relationship between video file swarms is also observed from the inter-swarm measurement. This inter-swarm relationship can provide extra peer information for video file swarms. Therefore, the organization of existing peer information could be the key to mitigate the preference bias. To this end, a hypercomplex based virtual channel mechanism is proposed. We find that, quaternions can provide efficient and meaningful approximation in the bitfield operations. Using quaternion based three space rotation, our approach can help peers to avoid preference bias during the data delivery. This is especially helpful at the beginning of the downloading since the peer only has few local pieces. More importantly, this mechanism is transparent and friendly to the existing BT swarms with both video and non-video contents.