Incorporating Random Linear Network Coding for Peer-to-Peer Network Diagnosis

Recent studies show that network coding improves multicast session throughput. In this paper, we demonstrate how random linear network coding can be incorporated to provide network diagnosis for peer-to-peer systems. We present a new trace collection protocol that allows operators to diagnose peer-to-peer networks. It is essential to monitor large-scale peer-to-peer applications by collecting measurements referred to as snapshots from the peers. However, existing solutions are not scalable and fail to collect measurements from peers that departed before the time of collection. We use progressive random linear network coding to disseminate the snapshots in the network, from which the server pulls data in a delayed fashion. We leverage the power of progressive encoding to increase block diversity and tolerate extreme block losses by introducing redundancy in the network. Peers cooperate by allocating cache capacity for other peers. Snapshots of departed peers can thus be retrieved from the network. We show how our protocol controls the redundancy introduced through progressive encoding and thus scales to large number of peers and tolerates high level of peer dynamics.