direct liquid cooling demonstration - center of expertise · direct liquid cooling demonstration...

19
Direct Liquid Cooling Demonstration Henry Coles Lawrence Berkeley National Laboratory Environmental Energy Technologies Division October, 29 2013– SVLG Data Center Efficiency Summit Palo Alto, California contact: [email protected] SVLG 2013 - Palo Alto, California

Upload: phamcong

Post on 21-Jul-2018

225 views

Category:

Documents


1 download

TRANSCRIPT

Direct Liquid Cooling Demonstration

Henry Coles Lawrence Berkeley National Laboratory

Environmental Energy Technologies Division

October, 29 2013– SVLG Data Center Efficiency Summit

Palo Alto, California

contact: [email protected]

SVLG 2013 - Palo Alto, California

1

SVLG 2013 - Palo Alto, California

Presentation

• Acknowledgments

• What is this technology?

• Demonstration Goals

• Setup

• Turning Data into Information

• Conclusions

2

Acknowledgements

• California Energy Commission – Paul Roggensack

• Asetek – Steve Branton, Anderson Tsai, Steve Empedocles

• Cisco – Roy Zeighami

• Intel – Winston Saunders

• Joulex (now part of Cisco) – Scott Paisley, Oliver Stӧneberg

• Syska Hennessy Group – Ian Kucma, Vali Sorrell

• ServerTech - Dave Bercovich, Bruce Auclair

• LBNL High Performance Computing Services – Gary Jung, Manager – Susan James – Yong Qin – Karen Fernsler – Bernard Li

• LBNL High Performance Computer Facility – Mike Magstadt

• LBNL Colleague – Steve Greenberg

SVLG 2013 - Palo Alto, California 3

What is it?

SVLG 2013 - Palo Alto, California

Direct Liquid Cooling

4

Direct Cooling of

CPU and Memory

CPU Cooling Concept

Asetek RackCDU D2C ™ Liquid Cooling

SVLG 2013 - Palo Alto, California

Demonstration Goals Estimate Energy Savings of This Technology Compared to A Base Case

5

SVLG 2013 - Palo Alto, California

Goal: Key Variables

• Supplied Water Temperature 15C - 45C (59F - 113F)

• Water Flow Rate (Return Water Temperature)

~0.5->4.9gpm = ~17C-~50C (62F – 122F )

• IT Heat Level ~100 w/server (idle) –> ~440 w/server (max. power)

6

Goal: Determine Fraction of Server Heat Captured

for a Broad Range of Conditions

SVLG 2013 - Palo Alto, California

Setup

7

38

Asetek

Cooled

Cisco

Servers

Lytron

CDU

Not Part

of Asetek

Product Back of Rack with

38 Cisco Servers

Asetek

CDU

Setup (cont.)

SVLG 2013 - Palo Alto, California

Supply Water Temp. Control Metering: Btu, Electrical Power

Power Meter

Btu Meter Lytron

CDU

8

SVLG 2013 - Palo Alto, California

Setup (data collection)

9

SVLG 2013 - Palo Alto, California

Collected Data (sample)

10

Turning Data into Useful Information

SVLG 2013 - Palo Alto, California

0

10

20

30

40

50

60

70

80

15 25 35 45 55

Per

cen

t o

f Se

rver

Po

wer

Wat

er C

oo

led

Rack CDU Return Water Temp. (C)

Asetek Cooling - Idle Server Load(117 watts/server, ~28C Inlet)

40C Water Supply

30C Water Supply

20C Water Supply

15C Water Supply

30

35

40

45

50

55

60

65

70

75

15 25 35 45 55

Perc

ent

of S

erve

r Po

wer

Wat

er C

oole

d

Rack CDU Return Water Temp. (C)

Asetek Cooling - ~50% Server Load(~270 watts/server, ~28C Inlet)

20C Water Supply

25C Water Supply

30C Water Supply

40C Water Supply

45C Water Supply

30

35

40

45

50

55

60

65

70

75

15 25 35 45 55

Per

cen

t o

f Se

rver

Po

wer

Wat

er C

oo

led

Rack CDU Return Water Temp. (C)

Asetek Cooling - 100% Server Load(~430 watts/server, ~28C Inlet)

45C Water Supply

40C Water Supply

35C Water Supply

20C Water Supply

15C WaterSupply

11

Data Center Models

SVLG 2013 - Palo Alto, California

Dry Cooler Case

12

2MW IT Load, N+20% Fixed Speed CRAHs (id. sel.), N+1 Water Cooled Chillers,

N+1 Counter-Flow Cooling Towers, N+1 UPS, 2N Step-Down PDUs

Base Case Cooling Tower Case

Chiller Only Case

SVLG 2013 - Palo Alto, California

Modeling Results: pPUE for Full IT Load

preliminary results

13

SVLG 2013 - Palo Alto, California 14

Modeling Results: pPUE Estimates for 50% IT Load

preliminary results

SVLG 2013 - Palo Alto, California

Modeling Results: pPUE Estimates for Idle IT Load

preliminary results

15

Conclusions

• This direct cooling technology should provide a significant reduction (~16-21%) in total data center energy consumed if implemented as modeled for an IT load of 50%.

• Using colder water when economically available reduces the total energy consumed for the climate modeled (San Jose CA) in our study. Other climates will provide different results.

• Significant site-level energy savings should be possible using this technology even though it does not capture 100% of the server waste heat.

SVLG 2013 - Palo Alto, California 16

Questions

SVLG 2013 - Palo Alto, California 17

Cisco

Roy Zeighami

[email protected]

Asetek

Steve Branton

[email protected]

Lawrence Berkeley National Laboratory

Henry Coles

[email protected]

SVLG 2013 - Palo Alto, California

Backup Slide Measured Air Cooled Server Power vs. Air Inlet Temperature

18

SVLG 2013 - Palo Alto, California

Backup Slide

Air Cooled Server Power Adjustment vs. Front Panel Air Inlet Temperature

19