Simple and Optimal Fault-Tolerant Rumor Spreading

We present rumor spreading protocols for the complete graph topology that are robust against an arbitrary number of adversarial initial node failures. Our protocols are the first rumor spreading protocols combining the following three properties: they can tolerate any number of failures, they distribute the rumor to all nodes using linear number of messages (actually they use strictly minimal $n-1$ messages), and if an arbitrarily small constant fraction $p$ of nodes (including the initiator of the rumor) are working correctly, our protocols communicate the rumor to all members in the network in $O(\log(n))$ rounds. Our protocols are simpler than previous fault-tolerant rumor spreading protocols in this model, they do not require synchronization (i.e., their correctness is independent on the relative speeds of the nodes), and they do not require a simultaneous wakeup of all nodes at time 0.