Disrupting the Connectivity of Multiagent Peering Networks: a Model Study

Abstract This report summarizes simulation results for the process of peering network disconnection caused by consecutive failures of agents. These failures are considered to be non-arbitrary, but rather forced by an external adversary who employs different prioritization schemes or ”attack scenarios”. In this report authors demonstrate the process of disconnection on such common random network topologies as random graph and Barabasi-Albert model. In order to evaluate the resilience of such network topologies to attack scenarios authors introduce two concepts (and, respectively, quantitative characteristics thereof): ”robustness”, understood as a number of steps an adversary must perform to disrupt the network, and ”intactness”, calculated effectively as a number of remaining routes between agents after an attack. Authors propose to use robustness metric to evaluate the overall network vitality under targeted attacks, and intactness to measure the effectiveness of disruption (or the direness of consequences of an attack). The model represented in this report simulates 5 types of adversarial prioritization: random node removal, removal by degree - either starting with the lowest or highest, removal by the clustering coefficient and by eigenvector centrality. The report demonstrates robustness and intactness neasures for each of these attacks against different types of topologies. Further research will be directed towards employing game-theoretic methods to analyze different adversarial strategies and to devise effective defence algorithms which will substantially increase multiagent network resilience.