Data plane programmability for dynamic service provisioning

With a steady progression toward network softwarization, the new generation of communication infrastructures have undergone drastic transformation in recent years. Due to this, networks may now support higher levels of automation, pervasiveness, flexibility, and efficiency. The newest generation of network services must confront with the increasing needs of different application contexts. For instance, it is frequently necessary to provide an extremely tight latency (e.g. IoT and machine-to-machine communications). In contrast to cloud computing, fog computing proposes an intermediary layer between network infrastructure and end users, wherein clusters of fog nodes deliver services in response to service requests. Fog nodes, in addition to computational needs to the network, can offer a highly dynamic ecosystem by joining or removing themselves from the Fog cluster. To handle the many service models in this scenario and the associated resource monitoring and allocation challenges, a service orchestration system is required. In this thesis, a DoS prevention service is designed and implemented through the P4 programming language and P4Runtime framework, and prepared for integration with a resource management and service provisioning layer placed between the service consumers and a Fog infrastructure. These tools enable the effective softwarization and programmability of the Data Plane of network devices. The complete system providing the DoS prevention service is described in detail in each of the implemented layers, and the results of a performance evaluation are presented.