Enabling Artificial Intelligence Analytics on The Edge

This thesis introduces a novel distributed model for handling in real-time, edge-based video analytics. The novelty of the model relies on decoupling and distributing the services into several decomposed functions, creating virtual function chains (V F C
model). The model considers both computational and communication constraints. Theoretical, simulation and experimental results have shown that the V F C model can enable the support of heavy-load services to an edge environment while improving the footprint of the service compared to state-of-the art frameworks. In detail, results on the V F C model have shown that it can reduce the total edge cost, compared with a monolithic and a simple frame distribution models. For experimenting on a real-case scenario, a testbed edge environment has been developed, where the aforementioned models, as well as a general distribution framework (Apache Spark ©), have been deployed. A cloud service has also been considered. Experiments have shown that V F C can outperform all alternative approaches, by reducing operational cost and improving the QoS. Finally, a migration model, a caching model and a QoS monitoring service based on Long-Term-Short-Term models are introduced.