In: IEEE Intelligent Systems, Vol. 19, No. 5, pages 32-38. September 2004.
Abstract: Improving mobile agents' survivability during agent server failures with unreliable underlying networks is challenging. This fault tolerance approach deploys cooperating agents to detect server and agent failures and to recover services in mobile-agent systems. Three types of agents are involved: the actual agent, the witness agent, and the probe. The failure-detection-and-recovery protocol employs a message-passing mechanism among these three kinds of agents. The authors discuss different failure scenarios and their corresponding recovery procedures. They also present stochastic Petri net models for their approach and discuss survivability evaluations based on simulations.
Keywords: mobile agent; witness agent; probe; witnessing dependency; fault tolerance; stochastic Petri net.