When management agents become autonomous, how to ensure their reliability?

Increasingly nowadays, networks are managed in a hierarchical, yet evolving to a distributed manner. The managed network is divided into sub-networks or domains that are managed more or less independently by autonomous agents. Once the failure of an agent is detected, it becomes even possible to have a further improvement by re-affecting the management tasks of the unreliable agent among the other agents in a way to ensure that the whole network continues to be reliably managed. This provides the property of graceful degradation to the distributed management system. The work presented in this paper provides a first step towards this interesting improvement. To ensure that the whole network is still managed even if a number of agents become unreliable, it is necessary to install a mechanism that continuously checks the reliability of the agents. When unreliable agents are detected, the management tasks that they have been performing are re-distributed amongst the other still-reliable agents. At some time in the future, the agent with the abnormal behavior might recover, for example following a human intervention, and the tasks that have been re-distributed on the other agents should be assigned back to the recovered agents