Distributed Data Diffusion in Finite Time in Decentralized Networks
Zuoen Wang, Jingxian Wu · 2018
Finite-time distributed data diffusion in a decentralized network is studied in this paper. The objective of distributed data diffusion is to disseminate local data at each node to all other nodes in the network without a central controller. We propose to achieve distributed data diffusion with a linear iterative information propagation algorithm. In the proposed algorithm, each node maintains and updates a state vector with size much less than the number of nodes in the network. In each iteration a node broadcasts its current state vector to all its neighbors. The state vector at a node is iteratively updated by using a linear combination of its own state vector and those from all its neighbors in the previous iteration. The algorithm converges when all nodes have a copy of the data from all other nodes in the network. We analytically identify the design parameters that guarantee the fastest convergence of the algorithm in diameter time, that is, the number of iterations required for convergence equals the network diameter. The optimum designs of square grid networks are studied by using the analytical results. The algorithm is efficient with guaranteed diameter time convergence, and it is scalable in that the total amount of data transmitted by each node throughout the iterative process scales linearly with the number of nodes in the network.