https://www.mdu.se/

Container-based virtualization has gained widespread adoption due to its lightweight, scalable, and portable deployment model, making it a key technology in modern distributed and cloud-native systems. However, its use in industrial domains remains challenging because such domains require temporal predictability, commonly referred to as real-time behavior. Standard container technologies do not inherently provide such guarantees, and containerized workloads may experience significant execution-time variability due to contention for shared hardware resources, including caches, memory bandwidth, and processor interconnects.

This doctoral thesis addresses these challenges by investigating mechanisms for improving the temporal predictability of soft real-time containerized systems through adaptive resource reservation. The thesis makes four main contributions: (i) a systematic literature survey of existing approaches to real-time container-based virtualization; (ii) the design and implementation of a Kubernetes-based orchestrator for the deployment and runtime adaptation of real-time containers; (iii) a Hierarchical Resource Orchestration Framework enabling multi-layer resource management and adaptation; and (iv) a virtualization framework for dynamic control systems that jointly adapts execution rates and resource allocations according to system dynamics.

Collectively, these contributions advance the state of the art in real-time containerized systems by improving temporal predictability through resource-aware orchestration and adaptive resource management, thereby facilitating the adoption of container-based virtualization in industrial applications.