Decentralized replication algorithms for improving file availability in P2P networks

Abstract

Being autonomous and scalable, peer-to-peer systems provide a paradigm for sharing files in the Internet. However, different from conventional structured replication systems like content distribution networks (CDN), peers in an unstructured P2P system may have different, sometimes low, online availability, and usually get only partial information about the resources of the system. Therefore, how to achieve good system level file availability by autonomous peers is an important goal in P2P replication systems. In this paper, we investigate decentralized and cooperative resource allocation algorithms in a class of P2P systems that provide replication service. We formulate this replication problem as an optimization problem, and propose several heuristic algorithms respectively. They include (a) a random algorithm, (b) a group partition algorithm that relies on peers' forming groups, and (c) a greedy search algorithm based on an estimated system-level file availability target. We compare and evaluate these algorithms by simulations, and observe that each of them has advantages depending on the system parameters.