Wireless sensor networks are applicable for sensing data from any given environment; in order to guarantee the survival and persistence of any network data, either external storage or robust distributed in-network storage mechanisms can be utilized. To guarantee data persistence, previous works have relied on external storage. However, robust distributed in-network storage methods have been shown to be more efficient for wireless sensor networks. Some reasons that make in-network storage more practical include: • For high-duty cycle applications, the data sensing rate is much higher than the rate sensor nodes can upload data to an external storage system. • In-network processing of data sensed in the network would be infeasible if all data must be stored outside the network.

Data recovery in distributed in-network storage pertains to the partial or full recovery of any data stored in the network. This thesis presents the following research-contributions to the theory and practice of data recovery for distributed in-network storage in wireless sensor networks. • We propose a decentralized (n, k)-MDS coding mechanism with minimum overhead in communication. Using a progressive data retrieval scheme, a data collector contacts only the necessary number of storage nodes needed to guarantee data integrity. The scheme gracefully adapts the cost of successful data retrieval to the number of storage node failures. • The issue of energy-efficient data recovery in wireless sensor networks, with the objective of minimizing energy expenditure and thereby maximizing network lifetime, is considered in detail. We formally prove the NP-hardness of finding the optimal set of source nodes and optimal routes for offline data recovery in general networks. Moreover, we devise an optimal polynomial algorithm for acyclic networks. • We consider data recovery with minimal energy costs as nodes fail in an unknown sequence. In particular, we propose algorithms to maximize the network recovery capacity—the total number of successful data recoveries until an insufficient number of nodes exist to reconstruct the original data. We propose an online polynomial-time algorithm, which is within a constant approximation ratio of the optimal oracle algorithm; the oracle has complete knowledge of failures patterns. We propose a more computation-efficient polynomial-time algorithm, which is also within a constant approximation ratio of the optimal in acyclic networks. Our simulation results show that this algorithm performs well even in general networks.