An architecture for a learning-based autonomic decision system

Since today's networks use traditional centralized management systems, the management became costly with the growth in the number of network equipments and available services. It became then clear that it was necessary to distribute the central management responsibilities throughout the network equipments. This article proposes an autonomic decision system with learning capabilities based on Artificial Intelligence concepts, to be used by network equipments in mesh-network environments without the need of a central knowledge. Considering a specific case of QoS routing, this system is able to establish bandwidth-aware communication paths between pairs of nodes in a distributed approach, without the complete knowledge of the network and the available bandwidth of its links. To develop this system in a real network an architecture has been specified and built, denote as DistArch. This architecture provides all the functionalities to perform distributed autonomic decisions in large-scale networks. The results of this approach show that, by cooperating with neighbour network agents, even without a full knowledge of the network state, it is possible to establish bandwidth-aware communication paths as optimal as the ones obtained with a central decision approach that contains full network information, being able to react to network changes with a fast convergence.