Constructing software autonomous agents to computer network management

Summary form only given. There are three important concepts in the computer network management area: managers, agents and managed objects. This work presents a methodology to develop autonomous agents for network management. There are two kinds of agents to develop: static or dynamic agents. The first one can be implemented, using heuristics obtained from an expert or the network administrator, through production rules or feedforward neural networks. Using the network examples we can construct dynamic agents. The recurrent neural network may be trained to solve a problem using some examples. Moreover, the behavior of the management must be considered, the network management may be reactive or proactive. From the analysis it is possible to define if the problem demands static or dynamic autonomous agents. The application of a static approach in the solution of dynamic problems cause two inconveniences: 1) it does not allow covering all the different states of a dynamic system; and 2) to cover all the states it would require a great neural network, perhaps impossible of converging to a solution.