Power-Law Chord Architecture in P2P Overlays Salma Ktari, Artur Hecker and Houda Labiod Telecom ParisTech, ENST France {ktari, hecker, labiod}@enst.fr ABSTRACT In this paper, we propose to extend the topology of Chord to a bidirectional graph to provide a super-peer based lookup algorithm and an efficient resource localization service at a very little additional cost. Keywords DHT, flooding, super peer, scale free network, DHT 1. INTRODUCTION Peer-to-peer networking systems consist of a large number of nodes or computers that operate in a decentralized manner to provide reliable global services, such as query resolutions or distributed computing. Existing P2P algorithms assume that all peers are uniform in resources. Messages are routed on the overlay without considering the differences of capabilities among participating peers. However, the heterogeneity nature of participating nodes is quite important. Therefore, it is possible to improve the performance of these algorithms by taking into account the heterogeneous nature of P2P systems. Several approaches propose to organize the topology through a two-level hierarchy by introducing the concept of super nodes (SN) [1]. The idea is to assign a large number of low-capacity nodes to one or more stronger node, which then becomes SN. The super-peer solution helps low capacity users. However, the design introduces non-uniformity, as an explicit hierarchy is imposed on nodes. Furthermore, the SN selection problem is highly challenging because in the P2P environment, a number of super nodes must be selected from a potentially big distributed system, in which neither the node characteristics nor the network topology are known a priori. In this paper we propose S-Chord, a new super-peer based lookup algorithm, without imposing a hierarchical architecture. The proposed algorithm uses a biased key- based routing and a power law like broadcast mechanism on top of a structured P2P network (Chord [3]). S-Chord search achieves a high query success rate and short routes, causing practically no additional cost. 2. SUPER PEER ROUTING IN S-CHORD 2.1 Motivation A scale free network [2] is a network whose degree distribution follows a power law (PL). That is, the probability that a randomly chosen node has degree k is given by P (k) = k - . In such networks, few nodes exhibit extremely high connectivity, while the majority is poorly connected. Scale-free networks exhibit many advantageous properties, such as small diameter (for 2 << 3, d ~ log log N), which allows fast searches, tolerance to random node deletions, and a natural hierarchy for an optimal usage of heterogeneous computing [4]. Such topologies can be exploited to provide scalable global services in highly dynamic and ad hoc environments. To identify the topologically important nodes that deal with large numbers of connections, we consider two metrics: the outdegree and the indegree for each node. We observe the fact that, although in conventional DHTs nodes maintain a well known number of outdegree neighbors (typically, outgoing degree of each node equals O(log N)); the indegree distribution is neither uniform nor normal. We measured the number of indegree links of each Chord node for a network size of 20000 and calculated the number of nodes corresponding to each indegree value. Figure 1 (a) shows the resulting indegree distribution (linear and log scale). From the chart, we can see that it is similar to a power law (PL) distribution especially for indegree values exceeding 30 (but a correct parameter estimation is work in progress). Indeed, few nodes exhibit extremely high connectivity, while the majority is poorly connected. As a result, Chord topology seems to be is a PL network, where most nodes have only log N neighbors, but some nodes have substantially more “contacts”. Nevertheless, Chord does not use its presumed PL knowledge for routing, but limits itself to log N outdegree neighbors (and ignores indegree neighbors). Therefore, it could be possible to design structured P2P networks that allegedly admit PL-like degree distributions from the conventional structured DHT. Our idea is to extend Chord to use its full neighborhood knowledge for routing and, if our assumption is correct, to establish a functioning PL network. Highly connected nodes may be considered as super nodes as they already have larger neighborhood knowledge. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CoNEXT 2008, December 9-12, 2008, Madrid, Spain. Copyright 2008 ACM 978-1-60558-264-1 ...$5.00