Data-Driven Autonomy in the Edge-Cloud continuum with application to smart cities

The increasing complexity of smart city infrastructures, characterized by large-scale, heterogeneous, and highly dynamic environments, poses new challenges to traditional approaches to distributed system management. The emergence of the Edge–Cloud continuum further amplifies these challenges by introducing a seamless integration of Cloud, multi-Cloud, Edge resources, Internet of Things, and human users, where services must operate under varying conditions of latency, resource availability, and network reliability. In such contexts, static configurations and centralized control mechanisms are no longer sufficient to ensure efficient system behavior. The concept of data-driven autonomy as a foundational paradigm for managing such distributed systems is investigated. Data-driven autonomy is defined as the capability of a system to dynamically adapt its behavior at runtime based on continuous analysis of system observability and contextual information, to maintain desired operational conditions in the presence of variability. Within this perspective, the continuum provides the appropriate abstraction for distributing computation, data, and control across multiple layers, enabling decentralized and adaptive decision-making. Adaptive resource management through a data-driven approach to microservice placement across the continuum is analyzed, enabling systems to respond dynamically to changing workloads and infrastructure conditions. The approach is investigated across multiple layers of the continuum, from centralized Cloud infrastructures to federated and Edge environments, where different constraints and requirements emerge regarding communication, resource availability, and locality. This progression emphasizes the need for telemetry-driven mechanisms capable of operating under increasingly heterogeneous and dynamic conditions. As the second autonomy dimension, the problem of Edge operational continuity is addressed by introducing a novel consensus protocol tailored to geo-distributed environments, designed to balance performance and availability, supporting meaningful local decision-making. Unlike classical consensus approaches, the proposed EdgeCloudWPaxos protocol leverages data locality rather than strict synchronization to adapt to unstable network conditions, thereby maintaining operational continuity across the Cloud–Edge continuum. Finally, context-aware optimization in smart city scenarios is investigated by leveraging data collected from both physical and human sources, enabling adaptive, context-aware decision-making at the city scale. Two representative case studies are considered: urban mobile sensing, where telemetry from municipal vehicle fleets is exploited to improve sensing coverage and data quality, and multimodal public transportation, where crowding-aware recommendations are used to balance passenger distribution across the network. This thesis demonstrates how data-driven autonomy can be achieved through a set of complementary mechanisms that address fundamental aspects of distributed system behavior. The proposed methods contribute to the design of smart city services, where continuous data flows enable systems not only to react to changes but also to autonomously optimize their behavior in alignment with their needs, thereby enabling self-adaptive Edge–Cloud systems that interact with the physical world. The effectiveness of the proposed approaches is validated through experimental evaluation and real-world use cases in smart city contexts.