Improving Reliability of Intelligent Agents for Network Management

The Intelligent Agent technology is being more and more used in treating all kind of information; in the case of network management, where machines monitor machines, a special care must be taken with respect to the reliability of the information. In the DIANA (Distributed Intelligent Agents for Network Administration)project, the main objective is to achieve network management using Intelligent Agents delegating tasks in a simple and easy way, using mainly the deliberative and reactive properties of agents [1]. The network manager that uses an Intelligent Agents system must trust his agents. He has to be sure all the time that each agent is running and that it has correct and updated information. Of course in most of the distributed agent systems this should be true, but within the network administration domain this is a condition sine qua non, since the information collected is used to control the network, to diagnose the connectivity and security states, to change configurations, etc. Faults in an Intelligent Agent system may occur: hosts going down, congested links, CPU overload, etc. and even the agent code may be corrupted. If an agent is not capable of accomplishing its goals, then either it must be replaced, repaired or the goals he has to achieve should be redistributed to the other agents. Hence, in order not to ruin the management system, the network manager and the system as a whole must be aware of the condition of its components, every possible faulty agent must be discovered as soon as possible, and the actual set of “faulty-free” agents must be known. We use the System-Level Diagnosis (SLD) to update continuously the knowledge of the reliability degree of the Intelligent Agent system. The SLD model was introduced in [3] to diagnose faulty units in a system. In a nutshell, SLD considers each