MAS & Fault-Management
Peng Xu, Ralph Deters · 2004
Despite the considerable efforts spent on researching and developing Multi-Agent Systems (MAS) there is a noticeable absence of deployed systems. In the past the MAS research community ignored this problem - arguing that it is not a genuine MAS problem and consequently of lesser importance than other unsolved issues like cooperation, coordination, negotiation and communication. However, as the field matures, empirical evaluations of techniques and systems are more commonly used and deployment issues like the management of a MAS become increasingly important. This paper introduces a generic framework for fault-management in MAS that has been successfully tested in a large scale MAS. 1. Motivation MAS are decentralized self-organizing systems consisting of autonomous entities called agents that are designed to solve tasks by cooperating with each other. Using structured messages for communication the agents negotiate and coordinate in a decentralized manner the actions required to solve a common task. Decentralization and self-organization ensure that there is no need for a central coordination and consequently no single point of failure resulting in a more robust design. But with the absence of centralized coordination it becomes difficult to determine the current state of the system or to predict the effects of actions. This difficulty concerning the determination and prediction of states is made worse by the fact that the functional dependencies between the agents change over time as a result of negotiations at run-time. Consequently MAS can appear to behave chaotically, which is one of the reasons for the lack of successful deployments. To overcome these difficulties it is necessary to introduce management functions that will support the monitoring and controlling of individual agents as well as the MAS. ISO/OSI defines five basic management functions for any distributed system namely: accounting management, configuration management, performance management, security management and fault management. While all five are important management functions we view fault management as the most relevant in the deployment of a MAS based on our experiences [1].