Understanding P2P-TV Systems Through Real Measurements Delia Ciullo, Marco Mellia, Michela Meo, Emilio Leonardi Politecnico di Torino, Italy, Email: {last name}@tlc.polito.it Abstract—In this paper, we consider two popular peer-to-peer TV (P2P-TV) systems: PPLive, one of the today most widely used P2P-TV systems, and Joost, a promising new generation application of which no previous measurement study has been considered. Besides the traditional measurements like the amount of generated traffic for signaling and data transmission, the novel contribution of the paper consists in investigating the content distribution mechanisms. In particular, we evaluate the characteristics of both data distribution and signaling process for the overlay network discovery and maintenance. By considering two or more clients in the same sub-network, we observe the capability of the system to exploit the locality of peers. We also explore how the system adapts to different network conditions. The methodology we develop allows also to identify periodic behavior of the application, highlighting bursts of both data and signaling traffic. I. I NTRODUCTION In 1997 and 1998, Francis and Zhang respectively asserted that IP multicast was going nowhere, and that some form of application-level multicast was needed to bring multicast to the masses [1], [2]. This primogenial proposal of end-user multicast relied upon the peer-to-peer paradigm and foresaw a tree-based end-host overlay architecture able to easily scale to hundreds of nodes. Offering multicast at the application layer seemed the most flexible solution, given the lack of scalability, violation of the routers stateless principle and some practical issues preventing the broad offer of multicast at the IP layer. The following years have witnessed the tumultuous diffu- sion of P2P systems for file sharing, with swarming mech- anism offering the best performance. The novelty of such systems relies upon two main ideas: cooperation among peers and peer resource sharing, i.e., sharing of content, bandwidth, memory and processing power. The new concept of P2P users providing service to other users, while obtaining services from the system, has several beneficial effects on performance: it in- creases system capacity, improves network service reliability, makes the network more flexible and adaptive to the users needs. These features and the key attribute of a limited server infrastructure cost naturally made P2P technology emerge as a strong candidate to satisfy the demand for live or near-live streaming over the Internet. Undoubtedly the very last period testifies the sprouting popularity of several P2P video stream- ing systems, such as PPLive [6], Coolstreaming, PPStream, TVants. They all have attracted millions of users to watch This work has been supported by the European Commission under the 7th Framework Programme STREP Project “Networks Aware Peer-to-Peer Application under WIse NEtwork” - NAPAWINE. live or on-demand video programs on the Internet [3], [4], [5]. Moreover a new generation of high-definition commercial P2P video applications, such as Joost [7], Babelgum, Zattoo, Sopcast, TVUnetworks, are at an advanced stage of proto- typing and beta-testing with the aim of capturing the IPTV business opportunities. These systems are targeted to offer large bandwidth video streams (1-10 Mbit/s) to a very large population of users (up to tens of millions). At the same time, the opportunity for P2P-TV systems to attract millions of users constitutes a worry for network carriers since the traffic they generate may potentially grow without control, causing a degradation of quality of service perceived by Internet users or even the network collapse (and the consequent failure of the P2P-TV service itself). Therefore, analyzing the behavior of P2P-TV systems and characterizing the traffic injected in the network are necessary steps toward the evaluation of the possible potential risks for the transport network. Unfortunately, the majority of above described systems are proprietary, and thus their protocols/architectures/algorithms are inaccessible. Some measurement campaigns of P2P-TV systems have been already presented in the literature, including both active and passive characterization of P2P-TV system and protocols, users behavior and service quality. In particular, a measurement study of PPLive is undertaken in [4]. Authors measured indexes such as bitrate, contacted peers versus time, geographic distribution of peers, etc. A similar characterization was performed in [8], in which authors analyzed the peer churning rate, the system stability and provided a more de- tailed description of control protocols used by the application. Both previously cited papers focus on PPLive and, to the best of our knowledge, no previous measurement about Joost is available. However, to date, no common methodology has emerged to characterize and understand different systems. In this paper we therefore propose a measurement methodology which allows us to obtain a detailed characterization of traffic generated by P2P-TV applications. In particular, by running several experiments in which one or more clients access to the P2P-TV service, we observe each local peer behavior under different network conditions. Simple indexes, like the down- load/upload bitrate, are first considered. Then more complex measurement indices are defined. In particular, looking at the peer connection pattern, i.e., the amount of traffic the peer sends to each other peer in the system over time, we identify regular and periodic patterns that could be difficult to highlight from traditional measurements.