Evaluation of P4P based on Real Traffic Measurement Y. Carlinet 1 , L. Mé 2 , Y. Gourhant 1 , H. Debar 3 (1) France Telecom R&D, (2) Supelec, (3) Telecom SudParis {yannick.carlinet, yvon.gourhant}@orange-ftgroup.com , ludovic.me@supelec.fr , herve.debar@telecom-sudparis.fr Abstract—Peer-To-Peer (P2P) traffic represents a significant proportion of the traffic today. However, currently widespread P2P systems take no account of the underlying network infrastructure. But by doing so, they would be able to perform a much more efficient peer selection. The P4P approach aims at allowing ISPs to cooperate with peers in order to improve their peer selection in the overlay. The objectives of this paper are to assess the benefits of P4P, both for the P2P system and the ISP, thanks to an experiment of a scale never achieved before, thereby complementing the results obtained in previous work. The results show that P2P applications need more information than just the Internet domain, in order to improve their performance, and that the inter-domain P2P traffic can be reduced by at least 72%. Keywords - Peer-to-peer, P2P, traffic optimization, traffic monitoring, transit/peering points, P4P. I. INTRODUCTION A. Background Peer-To-Peer (P2P) traffic represents a large proportion of the traffic at interconnection points [1]. Besides, in a P2P application, an overlay network is created so that the peers can share their resources, which are often duplicated among several peers. In this context, when a peer needs to access a resource, the selection is often performed at random, and it never takes into account the underlying network topology or congestion. A fortiori, P2P applications take no account of peering agreements between ISPs (Internet Service Providers), when performing peer selection. If P2P applications could have access to information about the underlying network, they would be able to perform a much more efficient peer selection. B. What is P4P P4P (Proactive Participation of Network Providers in P2P) is a consortium promoting evolutions to peer-to-peer applications so that they can better perform peer selection [14]. The aim of the P4P architecture is to allow network providers to provide information about the underlying network to the P2P applications in order to improve their peer selection. Thanks to this information, the P2P applications can select peers closer to them, with a better throughput or latency. The expected benefits for the ISPs are a better balance of the traffic load in their network (for instance to reduce traffic on a congested link), and a reduction of the traffic at interconnection points with other ISPs. The end-users will benefit from an overall better performance of their P2P applications (a better download rate for a file-sharing application for instance). C. Objectives of the paper The objective of the paper is to precisely evaluate the benefits of the P4P architecture with real users, based on a data trace from the operational network of France Telecom. To this end, we have setup a large-scale monitoring experiment in which we observed eDonkey traffic anonymously during around 10 months. We chose eDonkey because it is one of the most popular file-sharing applications in France. We have monitored a large sample of customers and we have recorded the files downloaded by them, and more importantly from which sources they are downloaded. In particular we have observed the AS (Autonomous System) the files come from. An AS is an administrative domain in the network, controlled by a single entity, typically an ISP. The experiment aims at counting the quantity of data that was downloaded by monitored customers, which could have been downloaded from inside the France Telecom AS. This way we obtain the quantity of traffic that could be saved at the interconnection points with other ASs, thanks to the P4P architecture. We also aim at evaluating the improvements in average download rate of customers when they download from peers close to them, inside their AS. The paper is organized as follows: Section II presents previous works which have proposed solutions for improving peer selection in P2P systems (II.A). Then it describes solutions for the P4P architecture and previous experiments performed to assess the expected benefits of the P4P architecture (II.B). Section III explains briefly the methodology employed to collect the data required for our study. Section IV details the data collected. Section V gives the results of the study about the expected benefits of P4P for ISPs, while Section VI deals with the benefits for P2P users. II. RELATED WORK The first part of this section focuses on the previous work in optimizing P2P systems with improved peer selection. The second part deals with the P4P approach. The Fifth International Conference on Internet Monitoring and Protection 978-0-7695-4023-8/10 $26.00 © 2010 IEEE DOI 10.1109/ICIMP.2010.25 129