Network coding for resilient peer-to-peer networks

Abstract

Peer dynamics (departure or failure) in peer-to-peer (P2P) networks disrupts the downloading process to the peers. Previous works on fault resilience include peer selection and overlay topology organization. Random network coding based on the generation concept is a practical approach to implement network coding for data networks in general. If we model the generations as blocks distributed by a seed in a P2P network and the disruptions caused by peer dynamics as packet losses, random network coding can be applied in P2P networks. In this paper, we propose a network coding scheme using a concept we call generation crossing. We also introduce a full cardinality precode, which enables a peer to reconstruct the original data file when only a subset of data blocks is received. We show that the success rate for the whole file to be delivered to peers increases remarkably with a small number of overlapped packets between successive generations. Computer simulations based on time-parametrized graph are presented. Simulation results confirm that network coding and generation crossing can provide resilience to peer dynamics in a P2P network.