Exception Handling in Agent-Oriented Systems

Agent-oriented programming may be the next generation paradigm to try and tame the software complexity beast. Agents are active objects capable of autonomous behavior. Mobility can be one of the attributes of agents in open systems. A software system could be structured as a dynamic, and possibly evolving, ensemble of cooperating agents. However, there is very little in the literature on how to effectively handle exceptions in agent-oriented software systems. Agent-oriented systems have all the exception handling concerns of sequential and concurrent systems, as well as some new issues that arise due to mobility and security in open systems. This paper develops an exception handling model whose salient feature is the separation and encapsulation of exception handling for an agent environment in a special agent called a guardian. The model presented here builds upon the notions of events, exceptions, notifications, and commands in an agent ensemble, and presents a number of exception handling patterns that can be used by a guardian. The model presented here is being investigated in the context of the Ajanta mobile agent programming system.