A Practical Network Coding Approach for Peer-to-Peer Distributed Storage

Abstract

In this paper, we analyze the performance of a peer-to-peer (P2P) distributed storage network based on the overlay architecture defined by the Wuala project. Although the original system is based on efficient erasure codes, e.g., Reed-Solomon codes, we investigate the use of a "simple" network coding strategy, which leverages on the well-known idea of randomized network coding. In particular, when a resource is published in the network, the fragments are randomly encoded with a predetermined overhead, which is consistent with the erasure coding strategy. Moreover, no regular network maintenance is scheduled to guarantee that a client node is able to successfully complete a resource download. We instead propose that a client node generates new fragments, to be stored in the network, when it is able to successfully download the entire resource. Our simulation results show that this simple coding strategy reduces the amount of data flowing in the network, thus obtaining more free disk space on storage nodes.