Genuine Atomic Multicast

This paper addresses the problem of atomic multicasting messages in asynchronous systems with unreliable failure detection. First, we give a characterisation of the notion of algorithms tolerating unreliable failure detection. Second, we give a characterisation of the notion of genuine atomic multicast. This characterisation leads to a better understanding of the diierence between atomic multicast and atomic broadcast , and to a clear distinction between genuine atomic multicast algorithms and non-genuine atomic multicast algorithms. Third, we show that in a system with at least three processes, the genuine atomic multi-cast problem cannot be solved with unreliable failure detection, if at least one process may crash. We mention however two ways of circumventing this impossibility result.