The cover time of neighbor-avoiding gossiping on geometric random networks

The standard gossiping used in many overlay networks consists in a Self-Avoiding Random Walk (SAW): a message, once received by a node, is forwarded to a node chosen uniformly at random among the neighbors, excluding the node it comes from. We focus on a generalization of the above walks, defined by the Neighbor-Avoiding Walks (NAWs), i.e. walks that not only avoid themselves, but preferably also the neighbors of the path they traveled. We studied the performance of NAW policies over geometric random networks (a common model used for unstructured networks for instance for Wireless Sensor Networks) in terms of cover time and as a function of several structural network graph metrics: nodes´ cardinality, nodes´ clustering coefficient, node distance distribution, link centrality distribution. We find that neighbor avoiding policies perform better that the usual SAW policy and that this improvement is especially apparent in networks whose topology is characterized by high values of link centrality.