Brief Announcement: Optimal Atomic Broadcast and Multicast Algorithms for Wide Area Networks∗

Distributed applications spanning multiple geographical locations have become common in recent years. Typically, each geographical site, or group, hosts an arbitrarily large number of processes connected through high-end local links; a few groups exist, interconnected through high-latency communication links. As a consequence, communication among processes in the same group is cheap and fast; communication among processes in different groups is expensive and orders of magnitude slower than local communication. Application data is replicated both locally, for high availability, and globally, usually for locality of access. In this paper, we investigate the atomic broadcast and multicast problems, two communication primitives that offer adequate properties, namely agreement on the set of messages delivered and on their delivery order, to implement data replication [3]. As opposed to atomic broadcast, atomic multicast allows messages to be sent to a subset of processes in the system. More precisely, messages can be addressed to any subset of the system’s groups Γ, each message possibly being multicast to a different subset. The set of groups to which a message m is multicast is denoted by m.dest. Atomic broadcast and multicast are defined by the primitives AXCast, i.e., A-BCast or A-MCast whether the message is broadcast or multicast, and A-Deliver. Moreover, we define the relation < on the set of messages that processes (correct or faulty) A-Deliver as follows: m < m′ if and only if a process A-Delivers m before m′. ∗The work presented in this paper has been partially funded by the SNSF, Switzerland (project #200021-107824). In the case of atomic broadcast, for every message m, m.dest = Γ.