Improving reliability of erasure codes-based storage paradigm under correlated failures for wireless sensor networks

In distributed sensor networks, ensuring data availability and reliability in the presence of node failures and malicious attacks is an important requirement. Traditionally, redundant schemes such as erasure codes and network coding are used to improve storage efficiency. However, prior works do not consider the scenario that node failures might cut the network into multiple components and result in unsuccessful data reconstruction. To address this problem, we first devise a data segment distribution scheme that enables randomly connected component of remaining network to have enough data symbols to recreate the initial data. Because the optimal symbol distribution is Nondeterministic Polynomial NP-complete problem, we further propose an approximation solution to solve it for arbitrary network model. Second, an efficient data recovery scheme with integrity check is proposed to reconstruct the initial data and repair the data saved on the disabled nodes in case of Byzantine failures. Compared with the previous approaches, the proposed scheme benefits from low data loss and storage overhead, which is confirmed by evaluations. Copyright © 2015 John Wiley & Sons, Ltd.