From edge to cloud: engineering the industrial compute continuum

The Fourth Industrial Revolution (Industry 4.0) fundamentally challenges traditional manufacturing systems and their established design principles. These traditional ISA-95-based systems struggle to meet the competing demands of modern production environments: real-time responsiveness on the shop floor and large-scale analytical processing in the cloud. Although the Compute Continuum promises to combine the advantages of cloud and edge infrastructures, the field still lacks a concrete engineering framework for its implementation. This research answers the central question: How can a unified cloud-native reference architecture be designed and implemented within the Compute Continuum to meet the changing demands of Industry 4.0? To achieve this, the work follows a multi-stage Design Science Research methodology. The process begins with a Systematic Literature Review to survey existing research and identify gaps. It then uses Attribute-Driven Design (ADD 3.0) to develop a new cloud-native reference architecture. Finally, it validates the proposed architecture through several quantitative and qualitative evaluations of its implementation and core management components. This dissertation organizes its main contributions as a six-stage engineering framework. This work first establishes a comprehensive Taxonomy of Compute Continuum Research, which systematically maps the state-of-the-art and identifies the research gaps that motivate the subsequent research. Then, it proposes the Cloud-Native Reference Architecture for the Compute Continuum in Industry 4.0 (CRACI), a decoupled, data-centric, cloud-native, and event-driven architecture that addresses the structural limitations of traditional industrial designs. Further contributions include: a quantitative analysis of the performance and resource trade-offs between primary implementation paradigms (e.g., different microservice frameworks) and deployment technologies (containers vs. unikernels); a message-based framework for the live migration of stateful services (MS2M); and two resource management frameworks, the centralized SAGA and the fully decentralized DREAMS, that use application-level semantics. The central argument of this dissertation is that realizing the Industrial Compute Continuum requires moving beyond generic, resource-centric management toward architectures that are application-aware and decentralized by design. The empirical evaluations validate this: Service Affinity demonstrably improves placement decisions over resource-only metrics, and message-based state reconstruction enables live migration of stateful services with minimal downtime. CRACI, MS2M, SAGA, and DREAMS together constitute a practical engineering framework for industrial systems that can scale, adapt, and operate reliably across the full continuum.