Cooling Effectiveness of a Data Center Room under Overhead Airflow via Entropy Generation Assessment in Transient Scenarios

Overview

The paper evaluates two strategies for delivering cold air into a data centre room with a dynamic modelling approach; the common method supplies air through perforated floor tiles from a centralised distribution system, which hinders flow management at aisle level. The idealised room has 10 racks of 10 servers (4U) each at a constant 1 kW per server, totalling 100 kW handled by five overhead heat exchangers, with a fully contained aisle and adiabatic room walls. In one case the exchangers force airflow downward into the aisle (ODF); in the other upward (OUF).

Servers and heat exchangers are compact differential-equation models in MATLAB (validated against experiments in earlier work) that include thermal mass, coupled to a transient k-epsilon Fluent model of the air side. The transient run reached steady state at t of about 3400 s, with the time step varied dynamically from 0.1 to 1 s. A mesh study found that only the finest mesh captured air-side entropy generation within 3%.

With containment, air-side entropy generation was only 0.15% (downward flow) and 0.03% (upward flow) of the total, compared with an earlier estimate of 34% for an uncontained legacy room. Overhead supply met less resistance than perforated-tile supply and gave a uniform distribution of cold air into the servers, with server temperatures uniform across the racks. Downward and upward flow gave nearly identical cooling performance, and a second-law analysis quantified the room cooling efficacy.

Limits: the geometry is idealised and the study gives no failure transients, so it is evidence for overhead and in-aisle cooling concepts and for compact-model coupling rather than for design values. The choice between upward and downward flow may depend on geometric, mechanical and economic aspects.

Key points

1. Idealised room: 10 racks of 10 servers (4U) each at a constant 1 kW per server, totalling 100 kW handled by five overhead heat exchangers, with a fully contained aisle and adiabatic walls.
2. Servers and heat exchangers are compact differential-equation models in MATLAB (validated against experiments in earlier work) coupled to a transient k-epsilon Fluent model of the air side.
3. The transient run reached steady state at t of about 3400 s, with the time step varied dynamically from 0.1 to 1 s; only the finest mesh captured air-side entropy generation within 3%.
4. With containment, air-side entropy generation was only 0.15% (downward flow) and 0.03% (upward flow) of the total, against an earlier estimate of 34% for an uncontained legacy room.
5. Overhead supply met less resistance than perforated-tile supply and gave a uniform distribution of cold air into the servers.
6. Downward and upward flow gave nearly identical cooling performance.

Using it in design

• Use in the design rationale that aisle containment cut air-side entropy generation to negligible levels (0.15% and 0.03%), against 34% for an uncontained legacy room.
• When comparing overhead or in-aisle heat exchanger concepts with perforated-tile supply, weigh that overhead supply gave less resistance and more uniform server inlet conditions; let geometry, mechanics and economics decide upward versus downward flow.
• Include server and exchanger thermal mass through compact models in transient cooling models, and label results from an idealised geometry as concept evidence, not design values.

Who should read it

Cooling designers and researchers evaluating overhead and in-aisle cooling concepts and transient thermal models.

Document contents

1. 1. Introduction
2. 2. Motivation and Goals
3. 3. Mathematical Model and Numerical Implementation (3.1 Airside, 3.2 Server Transient Model, 3.3 Heat Exchanger Transient Model, 3.4 Entropy Generation, 3.5 Numerical Procedure, 3.6 Grid Independence)
4. 4. Steady State Analysis (4.1 Entropy Generation in an Idealized Case)
5. 6. Conclusions

Read the document

Publication details

Authors

Luis Silva-Llanca; Marcelo del Valle; Alfonso Ortega; Andrés J. Díaz

Publisher

Entropy (MDPI AG)

Year

2019

Volume

21

Issue

1

Pages

98

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/e21010098

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation