Overlay construction strategies for peer-to-peer live streaming systems

Abstract

Video streaming now amounts to the majority of traffic on the Internet. Media streaming relies on large-scale content distribution networks (CDNs), that incur significant costs to build or use. Peer-to-peer distribution of video content reduces reliance on CDNs and costs.In peer-to-peer networks, peers share content in a topological overlay above the physical network. This is fundamental to networks performance. Unfortunately, peer-to-peer distribution is fraught with quality of experience (QoE) problems. For example, the simultaneous arrival of a large number of peers, known as flash crowd, can affect network topology and disrupt content transmission. In addition, in scenarios where users have limited bandwidth to contribute to the overlay, peer-to-peer systems need important mechanisms for peer contribution incentive in order to deliver media content for all peers.Existing peer-to-peer live streaming algorithms for constructing and maintaining network topology often face issues of high playback latency and media discontinuity problems. When peers achieve a larger number of partners, both control message overhead rises and sophisticated neighborhood filtering techniques are required to deliver media without disrupting the flow. Problems are particularly challenging when the percentage of free riders in the network is high, which is often the case. In order to deal with these challenges while mitigating free rider negative effects, we present the Peer Classification and Partnership Constraints (2PC) that constructs and maintains the network topology focusing only on simple peer-to-peer network configuration. The algorithm establishes the concept of peer classes in which peers are grouped by their media contribution to the network. These contributions are used to establish partnership criteria among classes, what we call Peer Partnership Constraints (PPC). Each of these classes sets up its peers with a limited number of out-partners in order to allow peers to send chunks in the request arrival order and avoid any sophisticated neighborhood filtering technique, significantly reducing system complexity. Moreover, constraints on peer classes prevent partnership competition between free riders and cooperative peers. This lack of partnership competition is fundamental to: (i) facilitate and speed up the process of new peers joining the overlay; and (ii) help 2PC improve playback latency and media discontinuity by bringing the class of the most contributing peers closer to the media server while moving free riders to the network edge (Silva et al., 2008; Liu, 2007). Our experiments show that 2PC ensures that the network can sustain 50% of free riders without disturbing cooperative peer partnerships on the overlay. 2PC requires low implementation complexity and in addition, incur on the low overhead of exchange messages on the network. Most important, since we are basing our solution on peer-to-peer configuration behavior, our strategies may be combined with current peer-to-peer approaches which face flash crowd events and which handle free rider peers.