thesis defense presentation summer 2015

37
Quantifying air flow rate through a server in an operational data center and assessing the impact of using theoretical fan curve Graduate Student : Neha Shigrekar Committee members: Dr. Dereje Agonafer (Committee Chair) Dr. Abdolhossein Haji-Sheikh Dr. Ratan Kumar Thesis defense Date: 07.28.2015

Upload: neha-shigrekar

Post on 13-Apr-2017

153 views

Category:

Documents


5 download

TRANSCRIPT

Page 1: Thesis Defense Presentation Summer 2015

Quantifying air flow rate through a server in an operational data center and assessing the impact

of using theoretical fan curve

Graduate Student: Neha Shigrekar

Committee members:Dr. Dereje Agonafer (Committee Chair)Dr. Abdolhossein Haji-SheikhDr. Ratan Kumar

Thesis defenseDate: 07.28.2015

Page 2: Thesis Defense Presentation Summer 2015

2

Agenda

• Motivation

• Objective

• Methodology

• Experimental work and Results

Page 3: Thesis Defense Presentation Summer 2015

3

Data center• A facility that centralizes an organization’s IT

operations and equipment

• Data is stored , managed and disseminated

• Servers are used as a medium of storage for the information on the internet

Page 4: Thesis Defense Presentation Summer 2015

4

Air movement in data center

Server inlet Server inlet

Server exhaust

Page 5: Thesis Defense Presentation Summer 2015

5

Current data center practices

• Maintaining lower temperatures to avoid hot spots

• More than one air handler to increase air velocity

• Neglecting the actual air flow rate across the server

• Air flow rate assumptions based on manufacturer’s data

Ref:- Datacenter 'Last Mile' Challenges- A Rack Perspective : S.Venkatraman, 10th September’09

Page 6: Thesis Defense Presentation Summer 2015

6

Why QUANTIFY the air flow rate?

Provision the servers with the right amount of air flow rate.

Utilize the conditioned air to efficiently cool the servers and in turn the racks.

Reduce power consumption caused by over provisioning the cold aisles with high velocity , low

temperature air.

Page 7: Thesis Defense Presentation Summer 2015

7

MOTIVATION

• Fans and servers are characterized in ideal data center conditions

• Operational set points are fixed based on the data derived

• Actual air flow across the server in an operating server is not quantified

Page 8: Thesis Defense Presentation Summer 2015

8

OBJECTIVE

• To study the impact of the operational data center conditions on the air flow rate across the servers

• To quantify the differences in the air flow rate in ideal v/s operational data center conditions

Page 9: Thesis Defense Presentation Summer 2015

9

Methodology

1. Characterize the fan unit for fan performance & server for server impedance per ANSI/ASHRAE Standard 51

2. Obtain the system operating points for a passive & active server per Standard 51

3. Quantify the system air flow rate at various fan duty cycles in the ideal conditions

4. Quantify the system air flow rate in an operational data center

Page 10: Thesis Defense Presentation Summer 2015

10

Air flow bench

Inlet

Static pressure tap

Differential pressure taps

Pressure transducers

Blast gateBlower

3000-5000 CFM Chamber

Page 11: Thesis Defense Presentation Summer 2015

11

Air Flow Bench

Nozzles Equipment attached to the inlet

Page 12: Thesis Defense Presentation Summer 2015

12

Ideal data center setup

hot

cold

Ref:- THERMO-MECHANICAL DESIGN CONSIDERATIONS AT THE SERVER AND RACK LEVEL TO ACHIEVE MAXIMUM DATA CENTER ENERGY EFFICIENCY- Dr. RICHARD EILAND

Page 13: Thesis Defense Presentation Summer 2015

13

Sunon Fan unit Fans in series used for system having high flow resistance

Ref:- http://www.nmbtc.com/fans/engineering/multiple-fan-use/

Page 14: Thesis Defense Presentation Summer 2015

14

Application of Fan Laws

• Fan units are placed in parallel to increase the air flow through the system

• – : experimentally derived

flow rate of a single fan unit

– : number of units within the system

Ref:- http://www.nmbtc.com/fans/engineering/multiple-fan-use/

Page 15: Thesis Defense Presentation Summer 2015

15

Theoretical Fan Curve

0 20 40 60 80 100 120 1400

0.5

1

1.5

2

2.5

1 FAN

Volumetric flow rate (cfm)

ΔP (i

n of

H2O

)

Page 16: Thesis Defense Presentation Summer 2015

16

System under consideration

• 1U HPSE1102 server

Fan units in parallel

Flow direction

InletExhaust

1U=1.719”

Page 17: Thesis Defense Presentation Summer 2015

17

0 10 20 30 40 50 60 70 80 90 1000

0.5

1

1.5

2

2.5

R² = 0.998543236143846

Volumetric flow rate (cfm)

System Impedance

Page 18: Thesis Defense Presentation Summer 2015

18

Operating PointΔP

(in

of H

2O)

0 10 20 30 40 50 60 70 80 90 1000

0.5

1

1.5

2

2.5

3f(x) = 0R² = 0

Practical operating pointPolynomial (Practical operating point)system impedance

Volumetric flow rate (cfm)

Theoretical operating point52.92 cfm

Practical operating point26.7 cfm

Page 19: Thesis Defense Presentation Summer 2015

19

Flow Curve

Ref:-Experimental and numerical characterization of a raised floor data center using rapid operational flow curves model, InterPACKICNMM2015-48234

Page 20: Thesis Defense Presentation Summer 2015

20

Considerations Re: Fan Laws

• There is assumed at least 1.5 to 2 fan diameter clearance

• In servers obstructions are placed in close vicinity of the fans

• HPSE1102 has a clearance of 0.5 fan diameter

Ref:- 2008 ASHRAE HANDBOOK- HVAC SYSTEMS AND EQUIPMENT; 2005 ASHRAE HANDBOOK- FUNDAMENTALS; ANSI/ASHRAE Standard 51

20mm

Page 21: Thesis Defense Presentation Summer 2015

21

Reasons

• Fan laws apply to aerodynamically similar fans running in same flow conditions

• The flow conditions for fans vary - Tested as a unit on the air flow bench with good i/o conditions -Mounted in a server with improper i/o conditions

Page 22: Thesis Defense Presentation Summer 2015

22

0 10 20 30 400

0.2

0.4

0.6

0.8

1

26.7

Passive Server

System ImpedancePolynomial (System Im-pedance)100%90%80%70%60%50%40%30%

Operating points at various duty cycles

Volumetric flow rate (cfm)

ΔP (i

n of

H2O

)

Page 23: Thesis Defense Presentation Summer 2015

23

Operating points at various duty cyclesΔP

(in

of H

2O)

Volumetric flow rate (cfm)

0 10 20 30 400

0.2

0.4

0.6

0.8

1

28.7

Active Server

System ImpedancePolynomial (System Im-pedance)100%90%80%70%60%50%40%30%

Page 24: Thesis Defense Presentation Summer 2015

24

Data center

Cold aisle Hot aisle

Page 25: Thesis Defense Presentation Summer 2015

25

Anemometers

Velocity anemometers attached in front of the servers

Anemometers plugged into Accusensor

ATM2400

Page 26: Thesis Defense Presentation Summer 2015

26

Calculation of air flow rate

• Velocity readings at different fan duty cycles

• Effective inlet area of the server considered

• Qwhere, - Q is the air flow rate - is the velocity -A is the effective area

Page 27: Thesis Defense Presentation Summer 2015

27

Volumetric flow rate vs Duty cycle

Volumetric flow rate (cfm)

Duty

cycl

e (%

)

2 4 6 8 10 12 140%

20%

40%

60%

80%

100%

120%

cfm vs duty cycles

13.28

Page 28: Thesis Defense Presentation Summer 2015

28

Observations• The air flow rate measured at same fan duty cycle in ideal data

center differs from the air flow rate in a data center by ~50%

• Fan laws cannot be applied to practical air flow rates to predict the air flow rate through a server in data center due to differences in flow conditions

• Flow conditions are altered due to presence of external impedances like presence of other servers, non-uniformity in the rack and pressure variations across the rack

Page 29: Thesis Defense Presentation Summer 2015

29

Summary of ResultsΔP

(in

of H

2O)

Volumetric flow rate (cfm)

10 15 20 25 30 35 40 45 50 55 600

0.2

0.4

0.6

0.8

1

1.2

13.28

26.7

52.92

Theoretical operating pointActive operating pointPassive operating pointOperational operating point

28.7

Page 30: Thesis Defense Presentation Summer 2015

30

Reasons • The operating conditions in the ideal data center are controlled -No pressure variations -No leakages -No bypass air flow -Uniform flow rate

• Operating conditions in the actual data center are hindered by - Non uniformity across the rack - External impedances -Variable pressures in both the aisles - Leakages in the rack - Bypass airflow

Page 31: Thesis Defense Presentation Summer 2015

31

Conclusions• A fan unit and server were characterized per Standard 51 and found to be in

agreement with the manufacturer’s published data

• The system performance was quantified in ideal data center conditions– The system operating point was found to be ~ 50% less than the theoretical operating

point

• The system performance was quantified in a representative data center environment– The system operating point was found to be ~ 75% less than the theoretical operating

point– The reason being the external impedances that cause pressure variations and thus affect

the air flow rate through the server

• This study has shown a significant difference in the observed air flow rate through the system as compared to the assumed operating point used by facility designers and operators to provision the cooling system

Page 32: Thesis Defense Presentation Summer 2015

32

Future Work

• Study the effect of aisle pressure on the fan performance in operational data center

• Consider the impact of non-uniformity within the rack

• Study the effect of clearances within the cabinet on flow uniformity

Page 33: Thesis Defense Presentation Summer 2015

33

AcknowledgmentsI am very grateful to…

Dr. Agonafer for his continuous support and encouragement Dr. Kumar and Dr. Haji-Sheikh for being a part of my committee Marianna Vallejo for mentoring me throughout my research Nirav Dodia, Ashwin Siddarth, Ruturaj Kokate and Divya Mani

for their support in experimental work Sally Thompson and Debi Barton for helping me in all the

educational matters My family and friends for all the moral support

Page 34: Thesis Defense Presentation Summer 2015

34

QUESTIONS

Page 35: Thesis Defense Presentation Summer 2015

35

THANK YOU

Page 36: Thesis Defense Presentation Summer 2015

36

Appendix1.Duty Cycle v/s RPM

2. Duty cycle v/s cfm passive server

3.Duty cycle v/s cfm active server

Passive operating point 100% 26.7 0.23483

Active operating point 100% 28.7 0.266659

Theoretical operating point 100% 52.92 1.09

Operational operating point 100% 13.28239 0.013

4. Operating points in different conditions

Page 37: Thesis Defense Presentation Summer 2015

37

Manufacturer’s Datasheet