Self-organized data redundancy management for peer-to-peer storage systems

Abstract

In peer-to-peer storage systems, peers can freely join and leave the system at any time. Ensuring high data availability in such an environment is a challenging task. In this paper we analyze the costs of achieving data availability in fully decentralized peer-to-peer systems. We mainly address the problem of churn and what effect maintaining availability has on network bandwidth. We discuss two different redundancy techniques - replication and erasure coding - and consider their monitoring and repairing costs analytically. We calculate the bandwidth costs using basic costs equations and two different Markov reward models. One for centralized monitoring system and the other for distributed monitoring. We show a comparison of the numerical results accordingly. Depending on these results, we determine the best redundancy and maintenance strategy that corresponds to peer's failure probability. © 2009 Springer-Verlag.