Energy efficiency of fog computing in access networks

Today’s and future delay-sensitive applications, such as connected vehicles, smart grids, smart cities, and Internet of things (IoT) applications generate tremendous amount of data that can overwhelm data centres in cloud computing infrastructures. Using these cloud data centres can cause a remarkable increase in the applications response time and in the power consumption. Thus, the efficiency of cloud computing is marred by such delay- sensitive applications due to the distance between the end-users and the cloud servers. To address these issues, fog computing units are placed in the access network to be close to the end-users and to offer processing and storage, resources. Passive Optical Network (PON) architectures are efficient technology choices in the access networks and data centres as their capacity provides high-bandwidth and their passive components offer reduction in the power consumption compared to electronic switching. In this thesis, we propose an energy efficient distributed fog computing architecture containing multiple fog nodes connected by a Wavelength Division Multiplexing (WDM) PON based on Arrayed Waveguide Grating Routers (AWGRs). The fog nodes can collaborate with each other through PON connectivity, and hence can meet the delay sensitive application requirements in access networks. Firstly, we develop a Mixed Integer Linear Programming (MILP) model to optimise the AWGRs connectivity between the distributed fog computing units to facilitate fog nodes collaboration through inter-fog communications. In addition to optimising the connectivity in the proposed collaborative architecture, we developed an energy-efficient resource allocation MILP model along with a heuristic model to optimise the placement of virtual V machines (VMs) while considering the inter-VMs traffic for fog units’ collaboration. The results show that optimising the placement of VMs in the distributed fog computing units saves up to 44% of the total power consumption compared to placing VMs while neglecting the power consumption and inter- VMs traffic. The impact of the volume of the inter-VMs traffic on the total power consumption is also investigated. In addition, we evaluate the collaborative proposed architecture capacity using a MILP model that optimises the placement of end-users’ demands while aiming to reduce the total networking and processing power consumption. The results show an increase in processing capacity utilisation and a reduction in the power consumption compared to a non-collaborative architecture. Finally, we study optimising processing user demands considering user-mobility which necessitate VMs migration between collaborative fog computing units over the WDM PON architecture. A MILP model is developed to minimise the total power consumption and propagation delay in case of conducting VMs migrations due to user mobility for delay-sensitive application. The optimal allocation results show that migrating the VMs over the AWGR-based PON achieves a significant decrease in networking power consumption and in propagation delay by 55% and 92%, respectively compared to using other routes with active components via the access network for the VMs migrations.