Gossip algorithms can be used for computing aggregation functions of local values across a distributed system without the need to synchronize participating nodes. Very recently, we have proposed acceleration strategies for gossip-based averaging algorithms based on ant colony optimization, which reduce the message and time complexity of standard gossip algorithms without additional communication cost. In this paper, we extend our latest studies by analyzing in detail how the proposed acceleration strategies influence the node selection of different variants of PushPull gossip algorithms and show that the directions of information dissemination across the network differ strongly according to the type of the underlying ``knowledge'' of the neighbors (local vs. global knowledge). This analysis leads to a better understanding of how information is spread throughout the network and provides important insights that can be used to further enhance the acceleration strategies.

Top