Intelligent agents for network management

A communication network is a set of nodes that are interconnected by links to permit the exchange of information. We model it as a network graph. A need to exchange information between two nodes is called a demand. Each demand requires a certain quality of service (QoS) when transferred through the network. The required QoS depends on the type of information to transmit and is constrained using several parameters. In order to satisfy a demand, we must allocate a route between the two endpoints of the demand satisfying the demand's QoS constraints. Assigning a route to a demand and reserving the resources needed is called establishing a circuit. Given a communication network, the problem is to allocate one route for each incoming demand in the network so that the QoS constraints of all demands are satisfied using the available resources of the network. We propose here a general framework using abstraction techniques based on blocking islands. The idea is to abstract the original communication network into a hierarchy of simplified graphs where each node, a blocking island, abstracts a part of the network inside which routing of demands requiring a given amount of bandwidth is possible. Each link of an abstract graph then identifies bottleneck links of the network. This hierarchy can be adapted dynamically in reasonable (polynomial) time to reflect the changes in the network's state, such as allocation or deallocation of circuits, link failures, or even network topology changes. The management of this hierarchy and the routing of demands can easily be distributed to intelligent agents.