Network Coding-based Data Storage and Retrieval for Kademlia

Abstract

Peer-to-peer distributed storage systems can be instrumental to develop solutions able to store the massive amounts of data generated by the Internet of Things (IoT) users. Given the higher probability of node failures, losses in the communication channels, and limited resources of devices compared to centralized storage solutions, it is key to minimize data retrieval time, while also maintaining high resiliency in the system. We propose a method based on random linear network coding (RLNC) for data storage and retrieval and the use of Kademlia for our peer-to-peer design to address these challenges. We analyze the performance of the proposed RLNC-based method theoretically as well as the traditional Kademlia in terms of data retrieval time and resiliency to node failures and channel losses. We use PeerSim to simulate the proposed method. Our theoretical analysis and simulation results show that the proposed RLNC-based method significantly outperforms traditional Kademlia for our core performance metrics. These gains in resiliency and data retrieval time are achieved while also reducing the data storage time for a wide region of operation. Our simulations show that only if the redundancy of the RLNC-based scheme is significantly increased (> 100 % redundant RLNC packets), then a small degradation (< 10 %) in data storage time occurs.