P2P TRAFFIC MEASUREMENTS ON THE EMULE SYSTEM Anna Puig-Centelles, Oscar Ripolles, Miguel Chover Dpto. Lenguajes y Sistemas Informáticos, Universitat Jaume I,. 12071 Castellón (Spain) {apuig, oripolle, chover}@uji.es ABSTRACT P2P systems are designed for the sharing of computer resources by direct exchange and they are characterized by their ability to adapt to failures and accommodate transient populations of peers while maintaining acceptable connectivity and performance. This paper examines the peer-to-peer file sharing infrastructure by focusing on a content delivery system: the eMule application based on the eDonkey protocol. We present a framework developed to collect data and to offer several measurements of the performance of this P2P network. This process will allow us to study the way clients use eMule with the aim of analyzing whether it is possible to create an overlaid network of communities which would enhance the performance of a P2P application. KEYWORDS Peer-to-peer, P2P, eMule, eDonkey, file sharing, content distribution. 1. INTRODUCTION Today, a significant part of Internet traffic is generated by peer-to-peer (P2P) systems. This new wave of network architectures is the basis for distributed computing systems. Such architectures are characterized by the direct sharing of computer resources rather than requiring the intermediation of a centralized server, offering a fault-tolerant and self-organizing capacity which needs an adaptive network topology. Our study focuses on those content distribution systems which are designed for sharing data between users. It is well-known that the traffic demands of P2P file sharing systems represent a significant percentage of the overall network traffic. The aim of this paper is to provide detailed traffic measurements over a P2P network which could be used to obtain a classification of peers into communities. Communities are created when one or more entities claim an interest in the same topic or have something in common (Puig-Centelles et al. 2007). The appearance of P2P systems and their increasing popularity in the last years, make the contributions of this paper more interesting, as these results can help in the modelling of future file sharing systems. In our case, we studied on the eMule file sharing application by using the eDonkey network. This paper is organized as follows: Section 2 presents the different P2P content distribution technologies and the previous work performed; Section 3 outlines the main aspects of the eMule application, as well as its protocol; Section 4 provides a detailed description of the framework we have implemented; Section 5 analyzes and discusses the results obtained considering important aspects of clients and files; Section 6 concludes our results and outlines future research. 2. P2P TECHNOLOGIES The operation of any P2P content distribution system relies on a network of computers (peers) and connections (edges) between them. This overlaid network is formed on top and it is independent of the underlying physical computer (typically IP) network. These virtual architectures can be characterized by the behaviour of their peers (Androutsellis-Theotokis & Spinellis 2004). Examples are shown in Table 1: Purely Decentralized Architectures (all peers are called servents because they act as servers and clients and there is