Energy-Efficient Function Scheduling via Prediction

Function-as-a-Service (FaaS) is a well-established paradigm in cloud computing, en- abling users to execute code in response to specific events without the need to man- age the underlying infrastructure. While widely adopted in cloud environments, the integration of FaaS with Edge Computing presents new challenges and opportuni- ties, particularly in terms of energy efficiency and Quality of Service (QoS). This thesis explores the application of FaaS in edge computing scenarios, with a specific focus on energy-aware function scheduling strategies that aim to meet functions’ QoS requirements. The work is organized in two parts. In the first part, we exam- ine a centralized scheduling approach, where function prediction models are used to optimize scheduling decisions. We evaluate the impact of different prediction mod- els on performance metrics such as energy consumption and resource utilization. In the second part, leveraging the inherently distributed nature of edge environ- ments, we propose and implement a distributed scheduling mechanism that allows nodes to make local decisions while still aiming for global energy efficiency and QoS compliance. Both approaches are thoroughly evaluated in terms of energy savings, with results showing that energy-aware scheduling significantly reduces energy con- sumption still ensuring that function deadlines are met. Specifically, the centralized strategy consumes only 8% of the energy, whereas the energy consumption of the distributed strategy ranges from 7% to 60%, depending on its configuration, but with a function drop of 11% and 15%, respectively.