As the number of smart meters/sensors increases to more than hundreds of thousands, it is rather intuitive that the state-of-the-art centralized information processing architecture will no longer be sustainable under such a big data explosion. Hence, an innovative data management system is urgently needed to facilitate the real-world deployment of a future residential distribution system. In this paper, we investigate a radically different approach through distributed software agents to translate the legacy centralized data storage and processing scheme to a completely distributed cyber-physical architecture. We further substantiate the proposed distributed data storage and processing framework on a proof-of-concept testbed using a cluster of low-cost and credit-card-sized single-board computers. Finally, we evaluate the proposed distributed framework and proof-of-concept testbed with a comprehensive set of performance measures.