Gossip-based counting in dynamic networks

Abstract. We propose Gossipico, a gossip algorithm to average, sum or find minima and maxima over node values in a large, distributed, and dynamic network. Unlike previous work, Gossipico provides a con-tinuous estimate of, for example, the number of nodes, even when the network becomes disconnected. Gossipico converges quickly due to the introduction of a beacon mechanism that directs messages to an au-tonomously selected beacon node. The information spread through the network shows a percolation-like phase-transition and allows informa-tion to propagate along near-shortest paths. Simulations in various dif-ferent network topologies (ranging in size up to one million nodes) il-lustrate Gossipico’s robustness against network changes and display a near-optimal count time. Moreover, in a comparison with other related gossip algorithms, Gossipico displays an improved and more stable per-formance over various classes of networks