Byzantine Fault Tolerance for Agent Systems

This paper presents a Byzantine fault tolerance method for agent systems. We extend Castro and Liskov's well-known practical Byzantine fault tolerance method for the server-client model to a method for the agent system model. There are two main differences between the methods. First, in agent systems we have to create replicas on both sides of the communicating agents, while in the server-client model of Castro and Liskov's method, replicas are created only on the server side, and the client is assumed to be non-faulty or is treated differently from a replica model. Second, due to the autonomous behavior of agents, we have to synchronize the timing of the receiving of messages among replicas. Agents decide their actions based on their current state of knowledge and do not wait indefinitely for messages that may not reach them