An assessment of Self-managed P2P Streaming Majed Alhaisoni Antonio Liotta Mohammed Ghanbari University of Essex Eindhoven University of Technology University of Essex malhai@essex.ac.uk A.Liotta@tue.nl ghan@essex.ac.uk Abstract Peer-to-Peer (P2P) IPTV applications have increasingly been considered as a potential approach to online broadcasting. These overcome fundamental client-server issues and introduce new, self- management features that help improving performance. Recently, many applications such as PPlive, PPStream, Sopcast, and Joost have been deployed to deliver live and Video-on-Demand streaming via P2P. However, the P2P approach has also shown some points of failure and limitations. In this paper we analyze, assess and compare two popular Live and Video-on-Demand P2P streaming applications, Sopcast and Joost. Our goal is not simply to underline similarities and differences from the architectural and functional point of view, but also to highlight their performance in terms of traffic behavior and network efficiency; thus our experiments have been conducted in different scenarios including a large-scale event (Euro 2008). . 1. Introduction Peer-to-Peer (P2P) video streaming is becoming a valid alternative to conventional IPTV systems for distributing video content over the Internet. The underlying mechanism is based on the distribution of the stream through a self-managed, application-level overlay including the user terminals in the role of peers i.e., content distribution relays. This is in contrast to other IPTV approaches which are based on content distribution networks. These require a dedicated multicasting infrastructure whose cost increases dramatically with the scale and dynamics of the system. On the other hand, in the P2P approach any capable (user) terminal becomes a distribution hub for any incoming stream, reducing in this way the possibility of failure points and bottlenecks that are traditionally associated with servers. Moreover, as the number of connected users increases, the number of distribution points grows too. As a consequence, the system scales much better than any client-server counterpart. The P2P streaming concept has nowadays been deployed into several trial P2P streaming systems, such as Sopcast [1], Joost [2], and Zattoo [3]. The online broadcasting arena is then evolving, mainly due to the clear commercial interest for these new technologies. However, despite most of the P2P applications are designed to automatically load-balance computing resources, they fail to pursue network efficiency. This article presents an experimental-based assessment of network efficiency in two of the most popular P2P applications that is Sopcast for the real-time broadcasting and Joost for the Video-on-Demand (VoD) services. By analyzing traffic traces we calculate and compare the network efficiency of these systems, in terms of network locality and percentage of P2P traffic (as a fraction of CS traffic). The percentage of P2P traffic, gives another form of computational efficiency since the aim of P2P systems is to minimize server intervention. We find, however, that due to the particular nature of P2P streaming it is not always possible to do without server support and that various systems address this issue in radically different ways. P2P traffic percentage gives an indication of both computational and network load balancing. We also look at the latter property from the view point of network locality, which is the ability to keep traffic local (in addition to being spread out). Our analysis reveals that network efficiency is being exploited poorly in the conventional P2P streaming systems, which indicate new opportunities for designing future applications aiming at network computational and cost efficiency. 2. Related Work Many studies have been published about P2P streaming, but very few actually focus on the analysis of the inter-relationship between computational and network efficiency deriving from self-managed, P2P networks. Existing work aims at understanding the underlying algorithms of current applications which are 2009 Fifth International Conference on Autonomic and Autonomous Systems 978-0-7695-3584-5/09 $25.00 © 2009 IEEE DOI 10.1109/ICAS.2009.47 34