An architecture for using modular redundancy to construct a computing node that is resilient to Byzantine failure of a minority of its constituent redundant computers is described. The architecture is based on a “perfect” communication bus, that is, one that will deliver all messages in a bounded time and will preserve at the receiver the order of messages at the sender. The correspondents of such a node communicate with it only via messages.
The basic idea is simple: no message leaves the node unless it has been agreed upon by a quorum of the redundant elements. It follows that input messages to such a node must be received by a quorum.
The interested reader must refer to the paper for the next level of detail. There he or she will find interesting distinctions between fail-signal and fail-silent modes of failure; an extensive discussion of the processes and data structures that support the architecture (called VOLTAN, for no explained reason); and an inventive technique for managing timeouts within the computers of the node in such a way as to preserve the necessary conditions for fault tolerance.