Dynamic Random Replication for Data Centric Storage

Abstract

This paper presents a novel framework for Data Centric Storage in a wireless sensor and actor network that enables the use of a randomly-selected set of data replication nodes which also change over the time. This allows reducing the average network traffic and energy consumption by adapt- ing the number of replicas to applications' traffic, while bal- ancing energy burdens by varying their location. To that end we propose and validate a simple model to determine the optimal number of replicas, in terms of minimizing av- erage traffic/energy consumption, from the measured ap- plications' production/consumption traffic. Simple proto- cols/mechanisms are proposed to decide when the current set of replication nodes should be changed, to enable new applications and sensor nodes to efficiently bootstrap into a working sensor network, to recover from failing nodes, and to adapt to changing conditions. Extensive simulations demon- strate that our approach can extend a sensor network's life- time by at least a 60%, and up to a factor of 10x depending on the lifetime criterion being considered.