Evaluation of Local Thermal Environment in Operational Data Center

Overview

When redeploying servers or installing new racks in an existing data centre, assessing whether target areas meet stringent cooling requirements is crucial, yet limited research addresses it. The authors built a 2U server simulator (675 W maximum, adjustable fans and heaters) and placed four in a test rack. The facility is about 650 m² with a 3.6 m ceiling, a 0.7 m plenum, eight uncontained rows (four cold and five hot aisles) and fifteen CRACs. Each 2U simulator measures 550 × 435 × 88.9 mm with three Delta PFB0812UHE fans, a PSU fan and six heaters; the rack was instrumented with 10 inlet and 30 outlet T-type thermocouples on frames 5 cm from the rack faces.

IOM is defined from the maximum and minimum rack inlet temperatures and the mean outlet and inlet temperatures, and values above 1 indicate exhaust recirculation; it picks up within-rack non-uniformity that RCI, SHI or RTI average away. With blanking, placing all four servers at the bottom of the rack gave the lowest IOM (0.28). Removing blanking plates raised IOM by 31%, 358%, 718%, 84% and 85% across the five layouts, and inlet temperatures at the same points rose by up to 29%; in two layouts IOM exceeded 2.0.

The local CFD domain was 2.2 × 1.8 × 2.15 m (three racks, 0.6 m aisles, 0.15 m overhead gap, a 0.05 m floor gap behind the rack); previous work shows three-rack models are more accurate than single-rack models. A three-rack local Icepak model, with servers as porous media calibrated to the simulator, reproduced the measured inlet profiles within a few percent: maximum inlet deviation was 3.31% in Case A, and 6.91% (inlet) and 8.61% (outlet) in Case B, with the largest outlet deviation 14.49%, attributed to surrounding hall effects and floor-gap leakage.

Limits: four servers under partial load and a single test rack; the local model does not fully capture the surrounding hall's influence and floor-gap leakage. The work supports local CFD sub-models plus simulator testing for brownfield deployments.

Key points

1. The facility is about 650 m² with a 3.6 m ceiling, a 0.7 m plenum, eight uncontained rows (four cold and five hot aisles) and fifteen CRACs; supply grilles are 0.6 × 0.6 m louvered (effective area 0.52 × 0.3 m, 60 degree discharge angle).
2. Each 2U simulator measures 550 × 435 × 88.9 mm with three Delta PFB0812UHE fans, a PSU fan and six heaters (675 W maximum); the rack had 10 inlet and 30 outlet T-type thermocouples.
3. IOM is defined from maximum and minimum rack inlet temperatures and mean outlet and inlet temperatures; values above 1 indicate exhaust recirculation. With blanking, all four servers at the bottom of the rack gave the lowest IOM (0.28).
4. Removing blanking plates raised IOM by 31%, 358%, 718%, 84% and 85% across the five layouts, with inlet temperatures up by as much as 29%; in two layouts IOM exceeded 2.0.
5. The local CFD domain was 2.2 × 1.8 × 2.15 m (three racks, 0.6 m aisles, 0.15 m overhead gap, 0.05 m floor gap behind the rack); three-rack models are more accurate than single-rack models.
6. Validation: maximum inlet deviation was 3.31% in Case A, and 6.91% (inlet) and 8.61% (outlet) in Case B; the largest outlet deviation was 14.49%, attributed to surrounding hall effects and floor-gap leakage.

Using it in design

• Before adding servers to an existing hall, measure the local thermal environment of the target rack in place with server simulators and score it with IOM; RCI, SHI or RTI average away within-rack non-uniformity.
• In a partly filled rack, place servers in the lower section (IOM 0.28) and close unused rack spaces with blanking plates; without panels IOM rose by 31-718%.
• For brownfield deployments, use a three-rack local CFD sub-model (servers as porous media calibrated to the simulator) with measurement validation, and allow for the largest deviation (14.49% outlet) coming from floor-gap leakage.

Who should read it

Operations and cooling engineers adding servers or rearranging racks in existing halls.

Document contents

1. 1 Introduction
2. 2 Methodology (2.1 Data Center Description, 2.2 Experimental Study, 2.3 Numerical Simulation)
3. 3 Results and Discussion (3.1 Effect of Server Layout and Blind Plates, 3.2 Rack Thermal Environment Simulation, 3.3 Local CFD Model Validation)
4. 4 Conclusion

Related standards

• ASHRAE TC 9.9 Thermal Guidelines for Data Processing Environments

Read the document

Publication details

Authors

Yan Xiao; Ruoyi Xie; Song Gao; Jinhui Yang; Luyao Liang; Shaozhi Zhang

Publisher

Frontiers in Heat and Mass Transfer (Tech Science Press)

Year

2026

Volume

24

Issue

4

Pages

1-10

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.32604/fhmt.2026.077396

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.techscience.com

Citation