Energy efficiency enhancement in two European data centers through CFD modeling

Overview

The EU's new reporting scheme focuses on assessing data-centre efficiency through key performance indicators (KPIs); the study asks how such KPIs can be calculated through CFD simulation and used to guide improvements. PSNC is an open-aisle micro data centre with 2 racks and 9 active servers cooled by ceiling-mounted units; AAU has 6 racks and 21 active servers cooled by two fan-coil units. AAU measurements used 16 calibrated Dantec 54R103 anemometers (0-5 m/s, ±5%) and PT100 sensors (setup uncertainty 0.21 °C); PSNC used IPMI data sampled every 10 minutes over two days.

Turbulence was modelled with k-omega SST. Servers were open-box models with momentum and heat sources and a Darcy-Forchheimer porous resistance (porosity 0.5), and cooling units were black-box fixed-flow boundaries with buoyancy from the ideal-gas law. Mesh3 was chosen after a five-level mesh study; final meshes had 1,652,180 cells (PSNC) and 1,066,351 cells (AAU) with y+ between about 1.1 and 9.8. Mean absolute differences between simulated and measured server inlet temperatures were 1.3 °C (PSNC) and 1.6 °C (AAU), with larger error near a server whose airflow is rotated 90 degrees.

KPI limits were ASHRAE recommended 18-27 °C and allowable 15-32 °C inlet temperatures. The PSNC retrofit (remove ceiling ventilation units, add separators to form hot/cold containment, seal unused server faces) gave a 50% RHI improvement and a 31% RI improvement. The AAU retrofit (reposition racks, separators, seal unused bays) gave a 75% RTI improvement and an 18% RHI improvement. RCI-HI can fall even when airflow is well organised if supply air is below the recommended minimum, so supply temperature and flow must be tuned separately from containment.

Limits: two small pilot rooms (2 and 6 racks), so results do not transfer directly to large halls. The paper presents the work as a KPI-guided retrofit workflow that fits the EU Energy Efficiency Directive reporting regime.

Key points

1. PSNC is an open-aisle micro data centre with 2 racks and 9 active servers cooled by ceiling-mounted units; AAU has 6 racks and 21 active servers cooled by two fan-coil units.
2. AAU measurements used 16 calibrated Dantec 54R103 anemometers (0-5 m/s, ±5%) and PT100 sensors (setup uncertainty 0.21 °C); PSNC used IPMI data sampled every 10 minutes over two days.
3. Turbulence by k-omega SST; servers as open-box models with Darcy-Forchheimer porous resistance (porosity 0.5); cooling units as fixed-flow black-box boundaries.
4. Mesh3 was chosen after a five-level mesh study: 1,652,180 cells (PSNC) and 1,066,351 cells (AAU), y+ about 1.1-9.8.
5. Mean absolute differences between simulated and measured server inlet temperatures were 1.3 °C (PSNC) and 1.6 °C (AAU); error was larger near a server whose airflow is rotated 90 degrees.
6. KPI limits: ASHRAE recommended 18-27 °C and allowable 15-32 °C inlet temperatures.
7. The PSNC retrofit gave a 50% RHI and 31% RI improvement; the AAU retrofit gave a 75% RTI and 18% RHI improvement.
8. RCI-HI can fall even when airflow is well organised if supply air is below the recommended minimum, so supply temperature and flow must be tuned separately from containment.

Using it in design

• Set up the retrofit workflow as KPI-guided: with a validated CFD model compute RCI, RHI, RTI and the recirculation index, then test rack repositioning, separators, sealing unused bays and removing badly placed ventilation units in sequence.
• Record the simulated-versus-measured server inlet temperature difference (here 1.3-1.6 °C) as an acceptance criterion for CFD validation, and expect higher error near servers with rotated airflow.
• When RCI-HI comes out low, tune supply temperature and flow separately from containment; supply may be below the recommended minimum.

Who should read it

Engineers preparing KPI-guided airflow retrofits and EU Energy Efficiency Directive reporting for small and medium server rooms.

Document contents

1. Introduction
2. Methodology (experimental studies at the pilot data centers)
3. Results and discussions (validation of the numerical model with the experimental data)
4. Conclusions

Related standards

• ASHRAE TC 9.9 Thermal Guidelines for Data Processing Environments
• EU Energy Efficiency Directive (EU) 2023/1791
• EU Code of Conduct on Data Centre Energy Efficiency

Read the document

Publication details

Authors

Onur Muhammed Sarikaya; Mustafa Kuzay; Ender Demirel; Simon Pommerencke Melgaard; Jesper Ellerbaek Nielsen; Rasmus Lund Jensen; Ariel Oleksiak; Wojciech Szeliga

Publisher

Scientific Reports (Springer Science and Business Media LLC)

Year

2025

Volume

15

Issue

1

Pages

24952

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.1038/s41598-025-11048-0

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.nature.com

Citation