Minimizing node churn in peer-to-peer streaming

Abstract

Several peer-to-peer (P2P) streaming systems have proved by now their ability to deliver live video streams to hundreds of users. However the inherent instability of the distribution environment poses several obstacles for these systems to manage to deliver a high quality experience to the end users. In this paper we explore node churn which independently of the distribution topology adopted, is an “anomaly” to the operation of the system leading to the degradation of playout quality. We argue that node churn is service specific and that churn in a P2P streaming service is highly correlated to the quality experienced at each node. On this basis we contribute a novel churn model to capture this twofold relationship and reveal unknown till now interactions while operating popular peer selection strategies under node churn. We provide evidence that selection strategies aiming solely at either efficiency or stability of a connection, although efficient for other P2P services such as P2P file sharing, lead to the formation of distribution topologies that are highly sensitive to node churn resulting in degraded performance. We propose a peer selection strategy designed to be P2P streaming service specific that takes decisions in short time scales while balances several factors such as connection efficiency, connection stability and content availability. It is proved that this approach achieves a uniform development of the distribution topology and leads to superior performance in terms of both low node churn and increased experienced quality.