SDN workload balancing and QoE control in next generation network infrastructures

The increasing demand of bandwidth, low latency and reliability, even in mobile scenarios, has pushed the evolution of the networking technologies to satisfy new requirements of innovative services. Flexible orchestration of network resources is increasingly being investigated by the research community and by the service operator companies as a mean to easily deploy new remunerative services while reducing capital expenditures and operating expenses. In this regard, the Future Internet initiatives are expected to improve state of the art technologies by developing new orchestrating platforms based on the most prominent enabling technologies, namely, Software Defined Network (SDN) orchestrated Network Function Virtualization (NFV) infrastructure. After introducing the fundamental of the Next Generation Network, formalized as the conceptual Future Internet Platform architecture, the reference scenarios and the proposed control frameworks are given. The thesis discusses the design of two resources management framework of such architecture, targeted, respectively, (i) at the balancing of SDN Control traffic at the network core and (ii) at the user Quality of Experience (QoE) evaluation and control at the network edge. Regarding the first framework, to address the issues related with the adoption of a logically centralized but physically distributed SDN control plane, a discrete-time, distributed, non-cooperative load balancing algorithm is proposed, based on game theory and converged to a specific equilibrium known as Wardrop equilibrium. Regarding the QoE framework, a cognitive approach is presented, aimed at controlling the Quality of Experience (QoE) of the end users by closing the loop between the provided QoS and the user experience feedbacks parameters. QoE Management functionalities are aimed at approaching the desired QoE level exploiting a mathematical model and methodology to identify a set of QoE profiles and an optimal and adaptive control strategy based on a Reinforcement Learning algorithm. For both the proposed solutions, simulation and proof-of-concept implementation results are presented and discussed, to highlight the correctness and the effectiveness of the proposed solutions.