A fully non-blocking reliable multicast protocol with total ordering

We present an efficient protocol for reliable multicast in an asynchronous network subject to link and process failures. Our protocol preserves total ordering in the sense that as processes or communication links become faulty, each group of non-faulty processes that remain connected will agree on the same sequence of messages delivered. Even processes that get disconnected deliver messages in a consistent order, i.e. message delivery is globally consistent. Although protocols that achieve reliable multicast with total ordering are known in the literature, these protocols will block the delivery of certain messages when the processes or links become faulty until membership assent is reached with the non-faulty processes. In contrast, our protocol is fully non-blocking in the sense that all messages will continue to be delivered by non-faulty processes despite a process failure or a change in membership of the multicast group. Moreover, our protocol differs from prior works in that it does not assume the existence of an underlying layer that detects link or process failures.