Prediction and Control of Airflow Distribution in RaisedFloor Data Centers

1
Prediction and Control of Airflow Distribution
in Raised-Floor Data Centers

• Suhas V. Patankar
• Professor of Mechanical Eng, University of
  Minnesota
• and
• President, Innovative Research, Inc.
• patankar_at_inres.com
• www.inres.com

2
A Raised-Floor Data Center
3
The Main Concern

• The flexibility of the raised-floor design offers
  an assured supply of cooling air wherever you
  place a perforated tile.
• Really?
• Does the air DISTRIBUTE in a uniform or desired
  manner?
• With complex layouts of computer rooms, CRACs,
  and perforated tiles, how can we know where the
  air goes?

4
Options for Managing Airflow

• Use Rules of Thumb
• A CRAC unit throws air up to 30 ft.
• Can you really depend on them?
• Conduct Measurements
• Expensive, time-consuming, and often
  impractical.
• Perform Trial and Error
• Expensive, time-consuming, and unreliable.
• Use a Predictive Model
• Easy-to-use, fast, accurate, and cost-effective.
• Based on CFD (Computational Fluid Dynamics).

5
Required Airflow Delivered
To CRAC Unit
75 F
75 F
75 F
2 kW
2 kW
300 CFM, 55 F
300 CFM, 55 F
6
Insufficient Airflow
To CRAC Unit
95 F
150 CFM
150 CFM
115 F
115 F
1 kW
1 kW
1 kW
75 F
75 F
1 kW
150 CFM, 55 F
150 CFM, 55 F
7
An Important Conclusion

• If the required cold airflow is delivered at the
  foot of each computer, proper cooling of the
  equipment is assured.
• If the needed cooling flow cannot be supplied at
  the perf tiles, the cooling of the computer
  equipment will be compromised.
• A good data-center design provides
• The correct amount of total airflow
• The desired DISTRIBUTION of airflow delivered to
  specific locations

8
Focusing on the Under-Floor Space

• It is is the fluid mechanics of the space below
  the raised floor that determines the distribution
  of the cold airflow through the perforated tiles.
  
• Thus, modeling of the flow in the under-floor
  space offers significant value for a modest
  effort.
• The computation is limited to the small space
  below the raised floor.
• The outcome is the valuable information giving
  the flow rate at each perforated tile.

9
Applications of the Airflow Model

• Initial Design. Ensure that the proposed layout
  gives the desired airflow for the immediate and
  anticipated needs.
• Failure Scenarios. Predict the behavior with
  failed CRAC units, redundancy, stand-by units,
  etc.
• Energy Savings. Are all CRAC units used to
  maximum benefit? Is the data center fully
  occupied?
• Changes in Layout. Data centers are evolving
  systems. Equipment is frequently relocated, new
  computers are installed, and new CRACs are
  commissioned. Before undertaking any actual
  changes, we should model the airflow distribution
  for the proposed layouts.

10
The Cause of Flow Maldistribution
CRAC
11
Data Center Floor Plan - Test Area
Work by Dr. Roger Schmidt, IBM Corporation
12
Calibrated Flow Tool
13
Boxed-In Test Area
14
Comparison With Measurements
15
Flow Maldistribution Revisited
CRAC
16
Effect of Plenum Height
(30 open tiles)
17
Effect of Tile Open Area
(12-inch plenum height)
18
Variable Tile Open Area Considered
CRAC
19
Use of Variable Tile Open Area
20
Achieving the Desired Airflow Distribution

• The airflow distribution is influenced by
• floor height
• CRAC geometry and locations
• perf tile layout
• perf open area
• under-floor blockages
• Whereas most blockages such as pipes and cables
  represent a nuisance, we can use under-floor
  partitions as a deliberate means of controlling
  the flow distribution in an imaginative manner.
• The use of perforated plates as partitions gives
  even greater control over the flow distribution.

21
Use of Perforated Partitions
Proposed Locations for Perforated Partitions
22
Perforated Partitions (70 and 30)
23
Comparison of Airflow Distribution
No Partitions
With Partitions (70 and 30)
24
Perforated Partitions (75 and 50)
25
Comparison of Airflow Distribution
No Partitions
With Partitions (75 and 50)
26
Perforated Partitions (80 and 65)
27
Comparison of Airflow Distribution
No Partitions
With Partitions (80 and 65)
28
Summary of the Exploration
Partitions (70-30)
No Partitions
Partitions (80-65)
Partitions (75-50)
29
Oakridge Data Center Layout
30
Oakridge Calculated Flow Rates
31
Oakridge Velocity and Pressure
32
Oakridge Comparison With Measurements
33
Oakridge Comparison With Measurements
34
Oakridge Comparison With Measurements
35
Oakridge Comparison With Measurements
36
Washington-DC Data Center

• In October 2001, extensive measurements were
  performed in a brand new data center in
  Washington, DC.
• The data center was designed using some new
  design concepts guided by the flow prediction
  model.
• The data center is about 10,000 sq.ft. in area,
  with 11 CRACs and over 200 perf tiles.

37
Washington-DC Calculated Flow Rates
38
Washington-DC Comparison With Measurements
39
Washington-DC Comparison With Measurements
40
Washington-DC Comparison With Measurements
41
Washington-DC Comparison With Measurements
42
Washington-DC Comparison With Measurements
43
Washington-DC Comparison With Measurements
44
Closing Remarks

• A computational model has been developed for the
  prediction of airflow through perforated tiles in
  a raised-floor data center.
• The model predictions have been validated using
  flow measurements in lab-scale and real-life data
  centers.
• The model can be used to predict the effect of
  tile perforations, raised-floor height, placement
  of CRACs, under-floor blockages, different
  layouts, failure scenarios, and so on.
• The use of the model can lead to data centers
  that are optimal, efficient, and reliable, and
  thus provide savings in capital and operating
  costs.