Dynamic routing in circuit-switched non-hierarchical networks
This thesis studies dynamic routing in circuit-switched non-hierarchical networks based on learning automata algorithms. The application of a mathematical model for a linear reward penalty algorithm is explained. Theoretical results for this scheme verified by simulations shows the accuracy of the model. Using simulation and analysis, learning automata algorithms are compared to several other strategies on different networks. The implemented test networks may be classified into two groups. The first group are designed for fixed routing and in such networks fixed routing performs better than any dynamic routing scheme. It will be shown that dynamic routing strategies perform as well as fixed routing when trunk reservation is employed. The second group of networks are designed for dynamic routing and trunk reservation deteriorates the performance. Comparison of different routing algorithms on small networks designed to force dynamic routing demonstrates the superiority of automata under both normal and failure conditions. The thesis also considers the instability problem in non-hierarchical circuit-switched networks when dynamic routing is implemented. It is shown that trunk reservation prevents instability and increases the carried load at overloads. Finally a set of experiments are performed on large networks with realistic capacity and traffic matrices. Simulation and analytic results show that dynamic routing outperforms fixed routing and trunk reservation deteriorates the performance at low values of overload. At high overloads, optimization of trunk reservation is necessary for this class of networks. Comparison results show the improved performance with automata schemes under both normal and abnormal traffic conditions. The thesis concludes with a discussion of proposed further work including expected developments in Integrated Service Digital Networks.