An autonomous network management architecture for hybrid communication systems

The relentless growth in wired/wireless communication technologies coupled with rapid advances in real-time applications renders the management of such technologies a major challenge. This dissertation investigates the Quality of Service (QoS) management problem and proposes a novel quadri-layer autonomous, self-adaptive and -reconfigurable management framework for hybrid wired/wireless communication systems. Necessary data pertaining to users, applications, domain and network components is collected at the first layer and utilized in the second layer to project future changes in the underlying network. These changes are fed to the third layer to reconfigure the underlying network components in order to best utilize the available resources while maintaining a smooth QoS delivery. Finally, the fourth layer provides a feedback mechanism based on network related measurements. This dissertation addresses the functionalities of these layers and describes novel schemes to realize those of the second and third layers.; To perform the functionalities of the second layer, a novel algorithm for mobility prediction through the utilization of Dempster-Shafer theory of evidence is developed. In contrast to existing approaches, the proposed scheme does not require a-priori training nor does it assume the availability of a history of previous users' movements.; The second component of the same layer presents a novel scheme for diagnosing probable network faults via statistical analysis and evidential reasoning aided with a training set of previously classified faults. The scheme also includes two new approaches to handle cases of imbalanced training sets.; Functionalities of the third layer are realized through the development of a new scheme for adaptive management of network-level QoS. The scheme addresses the management issue from a new perspective through posing it as a problem of learning from current system be havior while creating new policies at run-time. The scheme utilizes forecasting functions to estimate the impact of different adaptation decisions and to guide the decision-making process of adapting the behavior of network components.; Theoretical analysis and experimental studies using real data are presented to demonstrate the performance of the proposed schemes.