Analyzing the Impact of eDonkey Traffic on Internet Access Links Raffaele Bolla, Franco Davoli, Michele Sciuto DIST - Department of Communication, Computer and System Sciences University of Genoa, Italy f raffaele.bolla, franco.davoli, michele.sciutog @unige.it Abstract—Many measurement studies are showing that peer- to-peer (P2P) file sharing applications have evolved to become a major traffic source in the Internet. The ability to quantify their impacton the network is fundamental for network managers, who haveto guaranteea satisfactoryservicelevelto their customers. We present here a scalable simulation model, which aims at providing useful criteria in the capacity planning of Internet access links for enterprise networks, by considering the important effects of P2P file sharing traffic. Results are reported, along with a measurement study ofreal traffic traces,which confirms the validity of the model and its findings. I. I NTRODUCTION Nowadays, the success of P2P applications is evidently testified by the large amount of P2P Internet traffic carried by the Internet. Many measurement studiesrevealthatan important partof the total Internetraffic is carried by P2P protocols and that such a diffusion is not justrestrained to home broadband Internet accesses, but also involves enterprise networks[1] - [5] aswell asnetworksof industriesand commercial companies [6]. Network managers are nowadays paying attention to P2P traffic profiles in their networks. Such phenomenon increased the request of instruments not only for monitoring and measuring network traffic but also for modeling its behaviors. A relevant contribution in this direc- tion is represented both by analytical and simulation models, which can estimate the impact of traffic on the network layer, partially solving the difficulties still present in monitoring realtime traffic. Many modeling and simulation works look at P2P overlay networks, combined with the difficulties of simulating the large size overlay networks of real systems [7] - [13].In [8],a framework for P2P simulation environment, on top of existing packet-level network simulators, has been developed; the underlying layers seem to be considered in detail, even though this approach focuses on the packet level, which could not clearly scale to the size of a real P2P network. In [13] a survey of several P2P simulators revealed the lack of simulators, which could really be used for planning purposes. Well known scalable simulators are PlanetSim, Neurogrid and PeerSim,butthey have different aimswith respect to our proposal. In PlanetSim [10], developers can work at two main levels: creating and testing new overlay algorithms like Chord or Pastry, or creating and testing new services (DHT, CAST, 1-4244-1557-8/07/$25.00 c 2007 IEEE DOLR, etc) on top of existing overlays. Neurogrid [11] and PlanetSim can simulate large networks, but they are more oriented to the application layer, not being developed for dealing with network layer traffic. In general both scalability and network layer analysisare not supported features of existing P2P simulators. We present here a new scalable P2P simulation model, abl to represent up to 500,000 peers, with the aim of understan the P2P traffic impact on an Internet interconnection link of an enterprise network. We chose the eDonkey protocol as a reference system. We builtthe modelstarting from the overlay network and developing a file popularity and a query mechanism representation. We then included the presence of the network layer and the computation of the corresponding trafficmatrix.The modeldoesnot takeinto accounthe influence of the transport layer,butprovides an estimation of the offered load generated by P2P applications in an enterprise network. The division of files in multiple chunks is notconsidered, since the effects of such a feature on the offered load to a single Internet access link would notbe significant. In the lastsection we trace a future extension of the work, with a proposal for modeling the real throughput o P2P traffic, taking in consideration also the effects of lower layers. In the following we present our simulation model alon with measurement results performed in our campus network, with the purpose of finding a first general modelvalidation. Besides our measurements, we considered [14] as a relevant contribution, where a deep investigation of the real eDonkey network is reported. The remainder of this paper is organized as follows. Sec- tion II introduces the basic elements of the protocol and a description of the proposed simulation model. In Section III measurement results performed in our campus network are reported. Section IV presents simulation results in several scenarios, along with a model validation based on the measu ments reported in Section III. Finally,conclusions and future extensions of the work are shown in Section V. II. PROTOCOL AND MODEL DESCRIPTION In this section we introduce our model, which has the aim describing the P2P protocol and its effects on the network. In particular, we model 1) the overlay network, by considering its elements and its dynamics, 2) the query mechanism and the popularity of files and, finally,3) the network layer and 2007 Australasian Telecommunication Networks and Applications Conference December 2nd – 5th 2007, Christchurch, New Zealand 203