Density Classification in Asynchronous Random Networks with Faulty Nodes

This paper investigates distributed consensus for density classification in asynchronous random networks with faulty nodes. We compare four different models of faulty behavior under randomized topology. Using computer simulations, we show that (a) faulty nodes' impact depends on their location and (b) faulty nodes with persistent failures inhibit consensus stronger than commonly-used Byzantine faulty nodes with random failures. We also show that (c) randomization by Byzantine faulty nodes can be strongly beneficial for binary consensus and (d) topology randomization can increase robustness towards faulty node behavior.