A Progressive Fault Tolerant Mechanism in Mobile Agent Systems

We present the approach of deploying cooperating agents to detect failures as well as recover services in a mobile agent system. In additional to server failure detection, we use cooperating agents to handle agent failure detection. Two types of agents are involved. One is the agent performing the computation delegated by the owner, which we call the actual agent. Another is the agent that monitors the actual agent, namely the witness agent. We introduce a protocol by using a message passing mechanism between these two kinds of agents to detect agent failures and recover agent services. This approach can handle server failures, agent failures, and failures in message passing. It is capable of detecting and recovering most failure scenarios in mobile agent systems. We describe the design of our witness agent approach to mobile agent systems, and conduct reliability evaluation for our approach. The evaluation results show our approach is a promising technique in achieving mobile agent system reliability.