On Feasibility of P2P On-Demand Streaming via Empirical VoD User Behavior Analysis

In its current art, peer-to-peer streaming solution has been mainly employed in the domain of live event broadcasting. In such a paradigm, users are required to simultaneously participate the streaming, which yields tremendous bandwidth pool to alleviate the server load. However, little effort has been paid to study the performance gain when peer-to-peer solution is deployed into the domain of Video-on-Demand (VoD) applications, when users have the freedom to access a large pool of media files at their preferred times. Users of VoD applications exhibit file-dependent and time-varying access patterns, which are hard to simulate without realistic guidance from the operational system observation. In this paper, we present an empirical study on the traces collected by the Vanderbilt University media streaming service over a period of 8 months. We pay special attention to peer aggregation around one media file, in which peer-to-peer streaming is able to play an essential role. With this regard, we investigate three key factors: file popularity, request inter-arrival time, and user online duration. Our analysis proves the existence of skewed file popularity, concentrated user requests, and long enough online duration. Furthermore, through replaying the trace via simulation, we show that peer-to-peer streaming could reduce the server load by as high as 90% over popular files.