Bruno Donnassolo

Internet of Things (IoT) continues its evolution, causing a drasticallyegrowth of traffic and processing demands. Consequently, 5G players areeurged to rethink their infrastructures. In this context, Fog computingebridges the gap between Cloud and edge devices, providing nearby devicesewith analytics and data storage capabilities, increasing considerablyethe capacity of the infrastructure. However, the Fog raises severalechallenges which decelerate its adoption. Among them, the orchestrationeis crucial, handling the life-cycle management of IoT applications. Inethis thesis, we are mainly interested in: i) the provisioning problem,ei.e., placing multi-component IoT applications on the heterogeneous Fogeinfrastructure; and ii) the reconfiguration problem, i.e., how toedynamically adapt the placement of applications, depending oneapplication needs and evolution of resource usage.

To perform the orchestration studies, we first propose FITOR, aneorchestration system for IoT applications in the Fog environment. Thisesolution addresses the lack of practical Fog solutions, creating aerealistic environment on which we can evaluate the orchestrationeproposals. We study the Fog service provisioning issue in this practicaleenvironment. In this regard, we propose two novel strategies, O-FSP andeGO-FSP, which optimize the placement of IoT application components whileecoping with their strict performance requirements. To do so, we firstepropose an Integer Linear Programming formulation for the IoTeapplication provisioning problem. Based on extensive experiments, theeresults obtained show that the proposed strategies are able to decreaseethe provisioning cost while meeting the application requirements.

Finally, we tackle the reconfiguration problem, proposing and evaluatingea series of reconfiguration algorithms, based on both online schedulingeand online learning approaches. Through an extensive set of experiments,ewe demonstrate that the performance strongly depends on the quality andeavailability of information from Fog infrastructure and IoTeapplications. In addition, we show that a reactive and greedy strategyecan overcome the performance of state-of-the-art online learningealgorithms, as long as the strategy has access to a little extraeinformation.