From Fastnet Lighthouse to the Cloud: Dovetailing Resilience into Greener Data Centre Infrastructure

Containerisation has become a cornerstone of modern cloud infrastructure, offering efficiency, portability, and low overhead. However, its widespread adoption in cloud data centres raises concerns about energy consumption and environmental sustainability. While artificial intelligence (AI), and reinforcement learning (RL) in particular, has shown promise for optimising dynamic workloads, many existing solutions fail to consider the complexities of real-world deployments, where fault tolerance and continuous availability are crucial.

This research presents a novel lightweight AI-driven RL framework inspired by the dovetail toggle construction method used in the Fastnet Lighthouse, located off the coast of County Cork, between 1899 and 1903. The framework aims to enhance energy efficiency and ensure resilience in Kubernetes-based workflows in large-scale cloud data centres.

The Fastnet Lighthouse's interlocking granite blocks, secured through dovetail joints, provide structural integrity under extreme environmental pressure, enabling it to endure the Atlantic's harshest storms. This architectural technique is reinterpreted in the dovetail toggle migration pattern developed in this work, which interlocks node relationships through Kubernetes ReplicaSets, ensuring that workload scheduling and failure recovery are managed in an interconnected, energy-aware, and resilient manner. The pattern ensures that workloads are distributed in a way that anticipates instability and embeds resilience into the infrastructure, echoing the interdependent strength of dovetailed stonework.

The proposed approach dynamically adapts to fluctuating resource demands while maintaining high availability, using real-time energy telemetry to guide intelligent scheduling and scaling decisions. Each Kubernetes cluster functions independently, governed by a global RL agent. This agent coordinates optimal container placement and migration strategies based on the dovetailed architecture, and provides detailed recommendations for energy and CO2 savings in response to energy signals and resource imbalances.

By reducing RL overhead and enabling self-healing behaviour, the framework supports sustainable, fault-tolerant container orchestration through the targeted shutdown of underutilised nodes into cold standby mode. Experimental evaluation shows significant improvements in energy efficiency, CO₂ savings, and service continuity, with gains increasing at larger scales. These results outperform baseline Kubernetes scheduling, rule-based heuristics, and comparable state-of-the-art frameworks. While developed for container-based environments, the dovetail-inspired migration pattern and energy-aware placement strategy could also inform more efficient consolidation and standby approaches in traditional virtual machine infrastructures, offering a pathway towards greener, more resilient data centre operations.