| In this dissertation, the issues of long-term ATM network planning and short-term management are discussed. In order to formulate and solve the problems related to these issues, a layered framework based on the layer abstraction provided by ATM virtual paths is proposed. This framework includes long-term network planning, a short-term logical network reconfiguration, and real-time call management. The work of this dissertation focuses on the first two layers.;For long-term network planning, a genetic algorithm is proposed to obtain the minimum-cost physical network which meets the traffic requirements. The main advantage of genetic algorithms is that they converge to the global optimum. In contrast, most existing topology planning algorithms only converge to a local optimum. Our test results demonstrate that the genetic algorithm we propose can provide a very satisfactory solution, if not the best, in reasonable time if a suitable combination of crossover and mutation probabilities are used.;The objective of the short-term logical network re-configuration is to optimize the network performance. This problem is modeled both as a multicommodity flow problem, minDens, whereby the maximum flow density is minimized, and a nonlinear programming problem, minProb, in which the total blocked traffic is minimized. In both formulations, decomposition and coordination techniques are used. The outputs of both models are the traffic flows, which are used for the VP (Virtual Path) assignments. Because the number of VPs at an ATM switch is constrained by the maximum number of VPIs (Virtual Path Identification), a merge operation is proposed to make the VPs meet this constraint.;To evaluate the performance of the ATM networks designed and managed by the schemes proposed in this study, a flexible and object-oriented simulation platform is implemented. Our experimental results show that the algorithms we propose for ATM network planning and management are very effective. For the test network examined, to have a less than 10% blocking rate, the total planned network capacity should be more than 1.67 times of the total offered traftic.;To meet the challenge of satisfying the reliability requirements of high-speed ATM networks, formulations of the joint optimization of VP and restoration path assignments for a specific fault scenario and a survivable logical network design are proposed. A two-stage decomposition/coordination algorithm is designed to solve the problems. The solutions of the formulations are the assignments of the VPs and restoration paths. Due to the limit on the maximum number of VPIs at a switch, there may be an insufficient number of VPIs available to set up the restoration paths. Thus, a dynamic VPI assignment scheme is proposed, which assigns VPIs to the restoration paths for a link only when this link has a fault. Examples are presented to illustrate the details of the algorithm. |
