Spiral Walk on Triangular Meshes : Adaptive Replication in Data P2P Networks

📝 Abstract

We introduce a decentralized replication strategy for peer-to-peer file exchange based on exhaustive exploration of the neighborhood of any node in the network. The replication scheme lets the replicas evenly populate the network mesh, while regulating the total number of replicas at the same time. This is achieved by self adaptation to entering or leaving of nodes. Exhaustive exploration is achieved by a spiral walk algorithm that generates a number of messages linearly proportional to the number of visited nodes. It requires a dedicated topology (a triangular mesh on a closed surface). We introduce protocols for node connection and departure that maintain the triangular mesh at low computational and bandwidth cost. Search efficiency is increased using a mechanism based on dynamically allocated super peers. We conclude with a discussion on experimental validation results.

💡 Analysis

We introduce a decentralized replication strategy for peer-to-peer file exchange based on exhaustive exploration of the neighborhood of any node in the network. The replication scheme lets the replicas evenly populate the network mesh, while regulating the total number of replicas at the same time. This is achieved by self adaptation to entering or leaving of nodes. Exhaustive exploration is achieved by a spiral walk algorithm that generates a number of messages linearly proportional to the number of visited nodes. It requires a dedicated topology (a triangular mesh on a closed surface). We introduce protocols for node connection and departure that maintain the triangular mesh at low computational and bandwidth cost. Search efficiency is increased using a mechanism based on dynamically allocated super peers. We conclude with a discussion on experimental validation results.

📄 Content

Spiral Walk on Triangular Meshes : Adaptive Replication in Data P2P Networks. Nicolas Bonnel VALORIA European University of Brittany Vannes, France nicolas.bonnel@univ- ubs.fr Gildas Menier VALORIA European University of Brittany Vannes, France gildas.menier@univ- ubs.fr Pierre-Francois Marteau VALORIA European University of Brittany Vannes, France pierre- francois.marteau@univ- ubs.fr ABSTRACT We introduce a decentralized replication strategy for peer- to-peer file exchange based on exhaustive exploration of the neighborhood of any node in the network. The replication scheme lets the replicas evenly populate the network mesh, while regulating the total number of replicas at the same time. This is achieved by self adaptation to entering or leav- ing of nodes. Exhaustive exploration is achieved by a spiral walk algorithm that generates a number of messages linearly proportional to the number of visited nodes. It requires a dedicated topology (a triangular mesh on a closed surface). We introduce protocols for node connection and departure that maintain the triangular mesh at low computational and bandwidth cost. Search efficiency is increased using a mech- anism based on dynamically allocated super peers. We con- clude with a discussion on experimental validation results. Categories and Subject Descriptors C.2 [Computer Commmunication Networks]: Dis- tributed Systems General Terms Design, Algorithm, Experimentation Keywords peer-to-peer, replication, organization 1. INTRODUCTION In the last few years, the quality (reliability and band- width) of internet connections has increased drastically. With DSL, users have today an almost permanent high bandwidth access to the net. During the download of data, the transfert rate is now only limited by the ca- pability of the server to manage a large amount of high bandwith clients at the same time. Thanks to the peer- to-peer (P2P) mechanisms, each user can contribute to this data exchange mechanism as a potential server: this extends the sharing capability and features some fault resilient properties because of the data replication and distribution involved. Furthermore, P2P architectures allow users to share resources. Few existing architectures allow CPU cy- cle sharing [5], but most of them are designed to share memory and to behave as distributed databases. Data replication keeps information available when a node fails or leaves the network. Replication is useful for scalabil- ity because if a lot of users want to have simultaneous access to the same information, a single peer may not be able to supply all the queries. Several architectures have been proposed for P2P net- works. Napster [15], for instance, is a centralized P2P network that was very popular in the early 00’s. How- ever the use of a central repository to answer queries makes this system poorly scalable and vulnerable to failure. Many others P2P systems do not rely on a cen- tral server: they are decentralized, thus they are very scalable, and fault tolerant. Structured P2P networks associate network topology and location of data. Most of them implement a Dis- tributed Hashtable (DHT) [20, 16, 17, 23, 14] and pro- vide one basic operation: mapping a key to a specific node. This is performed by using a distributed hash function. Content routing is used to forward the key to the corresponding node. This strategy is very well suited for rare information retrieval (i.e. with a low number of replicas). Most of the structured P2P net- works can retrieve a key in O(log(N)) messages in a network containing N nodes. Their main limitation re- lies in poor performances for ranged and approximative queries, because hashing destroys the order on keys. Oceanstore [10], Cooperative File System (CFS) [4] or Glacier [6] are examples of storage systems, built on top of DHTs and on replicated data to provide high availability. Conversely, unstructured P2P networks do not fea- 1 arXiv:0805.4107v1 [cs.NI] 27 May 2008 ture any constraint between the location of data and the network topology. Gnutella [3], eDonkey 2000 [8], KaZaA [12] and Freenet [2] are examples of such work- ing architectures. The previous versions of Gnutella use flooding: this generates a number of messages that in- creases exponentially with the radius of the search. The trafic generated by this mechanism limits the scalabil- ity. To solve this problem, random walk, and variants such as k-random walk [13], have been proposed. If a node receives a query and cannot answer it, the query is forwarded to a randomly chosen neighbor. When a query is answered, it is then back-propagated to the node that initiated the query using the inverse path. Because random walk does not rely on the content of the query, much more complex queries can be processed, such as approximative or range queries. A Gnutella-like system has been proposed in [1]. The system relies on measures performed by [18] that show the heterogenity in capabilities of the peers involved in an unstru

This content is AI-processed based on ArXiv data.