Adaptive control of communication networks using learning automata
This research investigates communications network routing procedures, based on distributed learning automata concepts for circuit and packet switched networks. For this application, the learning automaton is shown to be an ideal adaptive control mechanism, with simple feedback and updating strategies which allow extremely practical implementations and perform very close to the desired optimum. In this thesis, the nature of learning automata routing schemes are explored by analytical and computer simulation techniques, primarily developing an elementary understanding of the automata routing and adaption process. Using simple circuit and message switched networks the conditions for minimum blocking probability and average delay are established and compared with the equilibrium behaviour of learning automata operating under alternative reinforcement algorithms. Later, large scale simulations of real networks are used to demonstrate and relate the learning automata scheme to existing routing techniques. These experiments, which are performed on sophisticated simulation packages produced for this study, take as examples hierarchical and general structured telephone networks and packet switched communications networks configured with both virtual call and datagram protocols. In addition, studies under failure mode conditions, including link, node and focussed overloads, conclusively demonstrate the superior performance afforded by the learning automata routing approach.