Hotspot management to reduce energy consumption in containerised data centres

The rapid increase in computing demand has increased the energy load of data centres, especially in containerised data centres where limited space worsens thermal problems. This thesis tests the hypothesis that the energy efficiency of containerised data centres can be improved by strategically distributing computational workloads to minimise thermal hotspots. To evaluate this theory, a computational fluid dynamics (CFD) model was developed based on the experimental containerised data centre system of Wang et al. (2017). A critical review of the literature established typical server characteristics, workload behaviours, and key energy metrics, highlighting significant gaps in validated thermal models and underscoring the need for realistic assessments of airflow, heat transfer and cooling performance in containerised data centres.
A parametric study of multiple Workload Distribution strategies was conducted to quantify their effects on hotspot formation, rack-level thermal behaviour, and cooling energy demand. Six workload configurations were analysed under steady-state conditions, complemented by a detailed examination of IT energy consumption across utilisation levels. The results demonstrate that baseline energy consumption remains substantial even at low utilisation, emphasising the importance of reducing idle capacity. Three workload strategies (cases D9, E12, and F15) successfully reduced hotspot intensity while maintaining cold-aisle temperatures within ASHRAE limits, resulting in observable reductions in total energy consumption. These findings indicate that coordinated workload allocation can enhance both thermal performance and cooling efficiency, although aggressive consolidation risks creating new hotspots that could offset potential energy savings. This research provides a validated framework for assessing the interaction between workload distribution, thermal behaviour, and energy consumption in containerised data centres. Although limited by steady-state modelling and simplified boundary conditions, the study offers a foundation for developing dynamic, environmentally aware scheduling algorithms. Future work should incorporate transient workloads, external climatic effects, variable airflow rates, and multiple cooling units to generalise the findings and support more energy-efficient containerised data centre designs.