An empirical study of data redundancy for high availability in large overlay networks

Redundancy is crucial for high data availability in an environment where computing nodes and/or communication channels are subject to transient failures, such as the Internet. Various techniques ranging from (multiple) replication to more sophisticated erasure correction coding have been studied. To the best of our knowledge, however, no easy rule of thumb has been devised to guide the distributed application designer in the choice of the appropriate technique and the right level of redundancy to guarantee the desired level of data availability. In this paper we present some simulation results that could help clarify the relation between the redundancy technique we adopted and the expected data availability in the context of medium to large size overlay networks, such as a Chord-like DHT.