International Conference on Automation, Control, Engineering and Computer Science (ACECS'14) Proceedings - Copyright IPCO-2014, pp.164-170 ISSN 2356-5608 Effects of Churn on Structured P2P Overlay Networks Zied Trifa a , Maher Khemakhem b a Department of Computer Science, University of Sfax, Tunisia b College of Computing and Information Technology, University of King Abdulaziz, Saudi Arabia a trifa.zied@gmail.com, b maher.khemakhem@fsegs.rnu.tn Abstract— As structured p2p systems becomes largely deployed, the size of the network becomes large enough that any kind of churn has a significant influence on the performance of such overlay. In structured p2p systems, peers create interactive link with other peers called neighbours that both depend on and influence each other. While peers trust each other’s to delivers services, the network effects exerted from peer to peer interactions remain virtually unexplored. This paper focuses on peer churn, where a significant number of peers join and leave the overlay frequently in a short time interval. The effects of such problem remain widely ignored yet may play a vital role in protection of such overlay. The direction and strength of this effect are evaluated experimentally on Chord and Kademlia protocol by measurement the set of metrics that describe the performance attributes such as efficiency, stability and scalability. Keywords— structured P2P overlay networks; churn; measurement; performance. I. INTRODUCTION Structured p2p systems, based on distributed hash table (DHT) such as Chord [1] and Kademlia [2], have the potential to harness huge amount of resource shared by peers having high connectivity. Unfortunately, it has been shown that most of these DHT systems suffer from a large fraction of churner peers, who refers to the effect of independent join and leave of a large fraction of peers in short interval time. The problem of churn, or the independent arrival and departure of participating peer, is a problem facing almost p2p networks. This phenomena is a major source of breaking down the overlay structure, because it is generally much more difficult to notify neighbours to restructure the overlay topology. Thus, causing stale neighbour pointers or data lose. Each instance of join or leave a network causes the protocol to rebalance its keys and the data among the present peers, which generate a considerable traffic load and degrades the performance of the system. Therefore, it is important to understand the effects of churn on the performance of such systems characterized by high dynamicity and decentralized nature. In structured p2p overlay networks, there is no barrier to join or leave the network. Peers can join or leave the network whenever they want. A peer joins the network when a participant starts the application and leaves when the participant exits. Although, such systems have to deal with churn problem. We define churn as the change in the set of nodes in the networks due to joins, friendly leaves or node failures. Join is such a condition that nodes join the network one by one or leave gracefully by informing their neighbour to restructure the overlay topology and save objects. However, a node failure is such a condition that large percent of nodes join or silently leave the network simultaneously and frequently [3]. This failure damages the structure and affects the performance of the overlay. In this paper we evaluate the impact of churn on the performance of Chord protocol as the most cited structured p2p overlay and Kademlia protocol as the most deployed one. The evaluation of such overlays is commonly done using defined scenarios and a set of metrics such as the number of hops, the response time, the number of requests and the success rate. The rest of the paper is organized as follows. Section II reviews the related works. In section III, we provide background information on Chord and Kademlia as DHT systems. Also, we introduce the problems of churn. Section IV describes the set of metrics and scenarios used in our measurement. In Section V, we give details analysis about the results of our measurement. Finally, section VI concludes. II. RELATED WORK Because the dynamic and decentralized nature of structured p2p overlay networks, join and leave process have been a prime issue from the beginning. Each DHT system employs a specific mechanism to deal with such phenomena. But after churning a large set of peers in a short time interval, the system will crashes. Unfortunately, very few studies examine the impact of churn problem. Abraham and al [4] study the scalability and resilience to worst case joins and leaves. They focus on maintaining a balanced network in the presence of high churn rate. Loguinov and al [5] examines graph-theoretic properties of existing peer-to-peer architectures and proposes a new infrastructure based on optimal-diameter de Bruijn graphs. They study routing performance of Chord, CAN and Bruijin. Li and al [6] present a comparative study of the effects of DHT parameters on the performance of structured p2p protocols under churn. Lam and al [7] address the question of how high a rate of node dynamics can be supported by these systems. They present a measurement studies to CAN protocol only. Furthermore, authors in [3] address the question of how to make p2p service available under churn and where is the