Autonomic Service Architecture for Next Generation Networks

Next generation networks will provide customers with a service mix placing variable demands for resources on the underlying infrastructure, motivating automated telecommunications services management approaches. This thesis proposes the Autonomic Service Architecture (ASA) for automated service delivery over next generation networks. First, we propose an architectural blueprint for ASA. We describe our view of the next generation network infrastructure, which will be application oriented. We elaborate on the layered design of ASA, the virtualization of resources, and the separation between manual and autonomic functions in the service delivery lifecycle. The autonomic functions are delivered by the interaction between Autonomic Resource Brokers (ARBs). The architecture of an ARB is then detailed, with a description of its different components and the message exchanges needed. Next, we discuss a Peer-to-Peer (P2P) naming and mobility management approach for next generation networks using ASA. This P2P approach will help ensure the scalability, robustness, and flexibility that ASA needs to ensure service delivery over next generation networks. The proposed P2P naming and mobility management infrastructure is then detailed, and its performance is evaluated. Finally, we suggest several autonomic resource management algorithms for ASA. The first algorithm is based on the Transportation Model, commonly used in the Operations Research community for cost minimization in delivering a commodity from sources to destinations, adapted to perform allocation of virtual resources. The second algorithm is based on the Assignment Model, commonly used in the Operations Research community for cost minimization in assigning several jobs to several workers, adapted to perform autonomic assignment of dedicated virtual resources. The third algorithm is based on Inventory Control, commonly used in the Operations Research community to analyze inventory systems, placing and receiving orders when needed for a given product, adapted to predict the demand on virtual resources. The fourth algorithm is based on Reinforcement Learning, commonly used in the Machine Learning community by agents to find a control policy that will maximize the observed rewards over their lifetime, adapted to adjust the prices of virtual resources.Ph