Predicting Availability of Mobile Peers in Large Peer-to-Peer Networks

Peer-to-peer (P2P) applications for mobile devices are becoming more and more popular because of increasing bandwidth, computational performance and storage capabilities. Such a mobile based distributed architecture offers significant advantages in several scenarios from the perspective of both users and network operators. In these situations, a certain redundancy must be built into the system because the availability of the nodes can vary greatly. The goal of this paper is to present a methodology to minimize the necessary redundancy by predicting the number of available nodes. The prediction is performed in two steps, first an estimation is done locally at the node level, then a centralized aggregation is executed. We propose a hypothesis which highlights the key advantages of our solution when compared to single step estimation. Finally we discuss about the correctness of the computed redundancy.