Design, Implementation and Optimization of Innovative Internet Access Networks, based on Fog Computing and Software Defined Networking

Abstract

1. DESIGN In this dissertation we introduce a new approach to Internet access networks in public spaces, such as Wi-Fi network commonly known as Hotspot, based on Fog Computing (or Edge Computing), Software Defined Networking (SDN) and the deployment of Virtual Machines (VM) and Linux containers, on the edge of the network. In this vision we deploy specialized network elements, called Fog Nodes, on the edge of the network, able to virtualize the physical infrastructure and expose APIs to external applications. Calling these APIs, external applications, are able to trigger and control the deployment of virtual environments, including applications, in form of VMs or containers, and SDN networking, directly on the access network. The main innovative points of this vision are: - usage of technologies developed for different fields, such as data center and IoT, in Internet access networks able to interconnect human users and smart objects; - considering the access network as a flat virtualized infrastructure, where it is possible to build the needed environments in real-time, including VMs, Linux Containers and SDN networks; - giving the control of the deployment of virtual environments, to external applications and in particular to the clients of the platform, through mobile applications (Apps), running on connected devices.

2. VALIDATION In order to validate the proposed approach, we developed a preliminary study focused on bandwidth optimization. We analyzed data collected on field in existing highly crowded Wi-Fi Internet access networks, in order to identify and measure the potential benefit of an infrastructure based on Fog Computing, compared to a traditional approach. Our results show that a significant portion (from 28% to 50%) of download data could be proactively managed by the Fog-based infrastructure.

3. FEASIBILITY In order to verify the feasibility of the envisioned platform and its compatibility with existing application, we deployed in laboratory, two testbed: one related to the Wi-Fi Internet access system, and another related to IoT applications. - Wi-Fi INTERNET ACCESS. We have developed the capability to trigger and control the deployment of full virtual environments, including networking and applications, directly at the edge of access networks, as a consequence of the interaction with a real authentication system, used to provide Wi-Fi public Internet access service. After the authentication process, we are able to deploy applications on Linux Containers and connect containers to user's devices and smart objects deploying a SDN network, according to the user's profile retrieved during the authentication. We measured that it is possible to add these new capabilities without a significant degradation of the network performance. - IoT HUB AND THE INTERACTION BETWEEN FOG AND CLOUD. Leveraging the virtualization capabilities of the Fog Nodes, and the interaction between the Fog and the Cloud, we introduced the concept of IoT Hub, hosted on the Fog, with replicas in the Cloud. We used this infrastructure, in order to prove the feasibility and advantage of the proposed architecture. In particular we show that this Fog-based application can run on devices usually used to manage Internet access networks, and that the interaction between Fog and Cloud improves security and reliability, without introducing delay that could affect performance.

4. BENEFITS EVALUATION IN REAL DEPLOYMENTS Redesigning provider's core applications, in order to leverage the presence of a Fog layer, it is possible to have great benefit. We investigated in particular the following aspects. - Authentication: a system based on the interaction between clients and the Fog Node is able to improve the users' quality of experience, measurable in a three times larger number of connections and users. - Location analytics: hosting on the Fog Node a virtual infrastructure able to locally process data provided by the Access Point infrastructure, we are able to collect and elaborate locally, more than 500,000 events every day, in a real location. We measured that this approach is able to produce real time analytics and daily report, saving 300 MBytes of traffic every day from the Node on site to the Cloud, compared to the system based on traditional approach. - Computing resources optimization: considering a set of 3,500 locations with Internet access service for guests, we measured how the nodes deployed on field are under-utilized, due to the highly variable level of workload. The nodes, already deployed for service management, have to be sized in order to manage the peak of workload, but they actually spend a big portion of their lifetime in idle. Redesigning the provider's core application for statistics, in oder to make it able to run on Fog Nodes, we show that it is possible to perform these tasks without involving the Cloud platform, but using the computing and memory resources already deployed for hotspot management, and underused in the traditional approach. The last point is particularly interesting, since we measured that, redesigning the provider's internal applications, it is possible to use the resources, already deployed on field, in order to perform the task usually performed at the data center. This analysis introduces the possibility to offer services in a Cloud-less approach leveraging the capability and flexibility of Fog.

5. ARCHITECTURE DESIGN AND OPTIMIZATION After the validation analysis, the description of the proposed platform, the description of specific use cases and real deployment, we conclude our dissertation focusing on the design and optimization of the architecture for the Fog-based Internet access network. Starting from the evolution of the role of access networks, from connecting end devices to hosting dynamic user virtual environments, allowed by Fog and SDN, we investigate the problem of the Fog layer design and optimization. We introduce the concept of static and dynamic overlapping clusters, and identify the latency between nodes as a key element that makes one approach better than others. Analyzing more than 89,500,000 measurements of latency between 4,116 Nodes in a nationwide network of Wi-Fi hotspots, we found that modern Internet connections provide performance able to support Dynamic Clustering of Fog Nodes, which is more flexible and adaptable. However, collected results also show that a not negligible rate of high latency Internet links are still active, forcing to organize some Fog Nodes in Static Clusters. For this reason, on a nationwide network, an hybrid approach should be preferred, using Dynamic or Static Clusters, depending on the type of the Internet connections.