Performance evaluation and modeling of active tile in raised-floor data centers: An empirical study on the single rack

Overview

Raised-floor data centres suffer local hotspots from uneven cool air delivery. The study examines a prototype active tile with eight 12 V fans under a standard tile and speed control from 1-100%. The test hall had a 4 m room height, 0.5 m plenum, 78 standard 42U racks around five open cold aisles of 0.61 m × 0.61 m tiles, underfloor supply with flooded-ceiling return, five 81.4 kW perimeter CRACs and a peak rack load of 6 kW. CRACs were set to 21 °C but delivered 23 °C supply, and thermal load was about 65% of cooling capacity, so one CRAC was off.

Rack inlet temperature profiles, tile airflow (using a 3×3 flow hood) and plenum-to-aisle pressure differential were measured at 11 tile settings, with and without the under-rack gap sealed. Tile pressure differential along a cold aisle was negative near the CRAC units and positive further away, because high-velocity discharge creates a low-pressure zone (Venturi effect). Replacing a passive tile with an active tile at only 10% fan speed raised mean tile velocity from 0.92 to 1.37 m/s, a 49% increase, while an unpowered active tile delivered only about 0.45 m/s against 0.95 m/s for the passive tile.

Active tiles lowered rack-top inlet temperatures by up to 4 °C but worsened the rack bottom, because low pressure at the tile surface drew hot air through the under-rack gap. Sealing the under-rack gap cut rack-bottom inlet temperature by as much as 6 °C (at 20% fan speed) and reduced inlet temperature variance from 5.19 to 0.42 at 20% fan speed and from 2.64 to 0.057 at 100%. Measured plenum-to-aisle pressure differentials were within about 1-4 Pa, and of the data-driven thermal models (GPR, ANN, MLR) GPR achieved the smallest prediction error, about 0.3 °C.

Limits: one data centre, a single rack and a prototype tile; results depend on this hall's geometry and CRAC layout. The authors stress that accurate thermal modelling of the active tile is the prerequisite for advanced control policies.

Key points

1. Test hall: 4 m room height, 0.5 m plenum, 78 standard 42U racks around five open cold aisles of 0.61 m × 0.61 m tiles with 55% porosity, five 81.4 kW perimeter CRACs, peak rack load 6 kW.
2. CRACs were set to 21 °C but delivered 23 °C supply, and thermal load was about 65% of cooling capacity, so one CRAC was off.
3. Tile pressure differential along a cold aisle was negative near the CRAC units and positive further away (Venturi effect).
4. Replacing a passive tile with an active tile at only 10% fan speed raised mean tile velocity from 0.92 to 1.37 m/s (49%); an unpowered active tile delivered only about 0.45 m/s against 0.95 m/s for the passive tile.
5. Active tiles lowered rack-top inlet temperatures by up to 4 °C but worsened the rack bottom, because low pressure at the tile surface drew hot air through the under-rack gap.
6. Sealing the under-rack gap cut rack-bottom inlet temperature by as much as 6 °C (at 20% fan speed); inlet temperature variance fell from 5.19 to 0.42 at 20% and from 2.64 to 0.057 at 100%.
7. Measured plenum-to-aisle pressure differentials were within about 1-4 Pa, and the GPR thermal model achieved the smallest prediction error, about 0.3 °C.

Using it in design

• With active tiles, always seal under-rack gaps; if left open, low pressure at the tile surface pulls hot air in under the rack and worsens the lower servers.
• Consider a fan-assisted tile for rack-top hotspots (up to 4 °C reduction), and remember an unpowered active tile adds flow resistance relative to a passive tile (0.45 instead of 0.95 m/s).
• Check in tile airflow design that a Venturi effect near CRACs can create negative plenum-to-aisle pressure differentials.

Who should read it

Cooling engineers and operators dealing with air distribution and hotspot removal in raised-floor data centres.

Document contents

1. 1 Introduction
2. 2 Related work (2.1 Passive tile, 2.2 Adaptive vent tile (AVT), 2.3 Active tile)
3. 3 Experiment setup (3.1 Measurement tools, 3.2 Measurement strategies)
4. 4 Experimental results and analysis (4.1 Rack inlet temperature, 4.2 Tile flow velocity, 4.3 Pressure differential)
5. 5 Data-driven models (5.1 Gaussian Process regression, 5.2 Artificial neural network approach, 5.3 Multivariate Linear Regression)
6. 6 Conclusion

Related standards

• ASHRAE TC 9.9 Thermal Guidelines for Data Processing Environments

Read the document

Publication details

Authors

Haoyu Gao; Qiuling Yue; Yuqing Kou; Jianxiong Wan; Leixiao Li; Lijun Fu

Publisher

Frontiers in Energy Research (Frontiers Media SA)

Year

2023

Volume

11

Pages

1073879

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3389/fenrg.2023.1073879

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.frontiersin.org

Citation