The use of mobile agent technology has been proposed for various fault-sensitive application areas, including electronic commerce and system management. A prerequisite for the use of mobile agents in these environments is that agents have to be executed reliably, independent of communication and node failures. In this article, we present two approaches improving the level of fault-tolerance in agent execution. The introduction of an itinerary concept allows to specify an agent's travel plan flexibly and provides the agent system with the possibility to postpone the visit of currently unavailable nodes or to choose alternative nodes in case of node failures. The second approach is a recently proposed fault-tolerant protocol to ensure the exactly-once execution of an agent. With this protocol, agents are preformed in stages. Each stage consists of a number of nodes. One of these nodes executes the agent while the others monitor the execution. After a summary of this protocol, we focus on the construction of stages. In particular, we investigate how the number of nodes per stage influences the probability of an agent to be blocked due to failures and which nodes should be selected when forming a stage to minimize the protocol overhead.
Lazy Agent Replication and Asynchronous Consensus for the Fault-Tolerant Mobile Agent System
