Control plane optimization in Software Defined Networking and task allocation for Fog Computing

As the next generation of mobile wireless standard, the fifth generation (5G) of cellular/wireless network has drawn worldwide attention during the past few years. Due to its promise of higher performance over the legacy 4G network, an increasing number of IT companies and institutes have started to form partnerships and create 5G products. Emerging techniques such as Software Defined Networking and Mobile Edge Computing are also envisioned as key enabling technologies to augment 5G competence. However, as popular and promising as it is, 5G technology still faces several intrinsic challenges such as (i) the strict requirements in terms of end-to-end delays, (ii) the required reliability in the control plane and (iii) the minimization of the energy consumption. To cope with these daunting issues, we provide the following main contributions. As first contribution, we address the problem of the optimal placement of SDN controllers. Specifically, we give a detailed analysis of the impact that controller placement imposes on the reactivity of SDN control plane, due to the consistency protocols adopted to manage the data structures that are shared across different controllers. We compute the Pareto frontier, showing all the possible tradeoffs achievable between the inter-controller delays and the switch-to-controller latencies. We define two data-ownership models and formulate the controller placement problem with the goal of minimizing the reaction time of control plane, as perceived by a switch. We propose two evolutionary algorithms, namely Evo-Place and Best-Reactivity, to compute the Pareto frontier and the controller placement minimizing the reaction time, respectively. Experimental results show that Evo-Place outperforms its random counterpart, and Best-Reactivity can achieve a relative error of <= 30% with respect to the optimal algorithm by only sampling less than 10% of the whole solution space. As second contribution, we propose a stateful SDN approach to improve the scalability of traffic classification in SDN networks. In particular, we leverage the OpenState extension to OpenFlow to deploy state machines inside the switch and minimize the number of packets redirected to the traffic classifier. We experimentally compare two approaches, namely Simple Count-Down (SCD) and Compact Count-Down (CCD), to scale the traffic classifier and minimize the flow table occupancy. As third contribution, we propose an approach to improve the reliability of SDN controllers. We implement BeCheck, which is a software framework to detect ``misbehaving'' controllers. BeCheck resides transparently between the control plane and data plane, and monitors the exchanged OpenFlow traffic messages. We implement three policies to detect misbehaving controllers and forward the intercepted messages. BeCheck along with the different policies are validated in a real test-bed. As fourth contribution, we investigate a mobile gaming scenario in the context of fog computing, denoted as Integrated Mobile Gaming (IMG) scenario. We partition mobile games into individual tasks and cognitively offload them either to the cloud or the neighbor mobile devices, so as to achieve minimal energy consumption. We formulate the IMG model as an ILP problem and propose a heuristic named Task Allocation with Minimal Energy cost (TAME). Experimental results show that TAME approaches the optimal solutions while outperforming two other state-of-the-art task offloading algorithms