2. Center Vision: To create electronic
systems that are self sensing and
regulating, and are optimized for
energy efficiency at any desired
performance level.
E3S works in partnership with
industry and academia to develop
systematic methodologies for
operating information technology,
telecommunications, and electronic
systems and cooling equipment
3. E3S is part of the NYS Center of Excellence in Small
Scale Systems Integration and Packaging
New electronics that improve the way people interact with their surroundings
E3S CASP
Energy-Efficient IEEC CAMM
Autonomous Solar
Electronic Systems Electronics Packaging Flexible Electronics
Supportive Analytical and
Diagnostics Infrastructure
5. Binghamton University is a leader in microelectronics R&D in
collaboration with government and industry. New $30 million Center
of Excellence building will house I/UCRC in Energy-Efficient
Electronic Systems, including new data center research laboratory.
6. Binghamton University
Bahgat Sammakia, Kanad Ghose, Bruce Murray
The University of Texas at Arlington
Dereje Agonafer
Villanova University
Alfonso Ortega, Amy Fleischer
Additional Collaborators:
The Georgia Institute of Technology
Yogendra Joshi
7. History of successful partnerships with industry
Research collaborations have 10+ year history
Faculty are leaders in this research effort with
expertise in:
◦ Thermal management techniques at all levels: chips to
racks to entire data centers
◦ Fast models for thermal analysis and trend prediction
◦ Scalable control systems for co-managing IT and cooling
systems
◦ Sensing-controls and model adjustments as needed
8. Endicott Interconnect NYSERDA
Technologies Advanced Electronics
Corning, Inc. Commscope
IBM
Sealco
General Electric Verizon
Emerson Network Power Comcast
Panduit
Steel Orca
GE DVL
Microsoft Microsoft
Bloomberg
9. E3S Members
Emerson
DVL
Facebook Collaborating Partner
10. Lifecycle: 90% of energy expended
during the operations
Breakdown of operating costs:
Others Equip
5% 45%
◦ Equipment energy consumption: 45%
◦ HVAC: 50% (about 25% to 30% in
water chilling facilities) HVAC
◦ Remainder: 5% Others 50% CPU
Breakdown inside a modern server: 28% 22%
◦ CPU energy: 20% to 25%
◦ Memory energy: 30% to 45%
◦ Interconnection energy: 10% to 15%
Total energy consumptions:
Interconn
12% Memory
◦ 2.8% of total national electricity 38%
consumption in US
◦ 1.6% for Western Europe
11. Server overprovisioning is a common practice
100%
Power
Performance is not proportional to
Energy
power
Efficiency
Typical
Poor energy efficiency
Operating
Region
(performance per Watt) on average
0% Individual Server 100%
Utilization
Situation becoming worse with smaller form
factors
12. HVAC equipment typically use water chillers to supply cold
air: takes a long time to adjust cooling system output
(water has a very high thermal capacity)
-Difficult for cooling system to track instantaneous load
changes
-Overprovisioning is common here as well to permit handling of
sudden load surges
-Using sensed temperature to adjust cooling
system performance is a viable first step,
but not enough
-Solutions for improving the energy efficiency of
data centers have been fairly isolated
System-level, holistic solutions are a MUST
13. Addresses energy-efficiency improvements of computing
and telecommunications systems at multiple spatial and
temporal scales:
◦ Synergistic techniques for simultaneously controlling IT load
allocation and thermal management in real time, to minimize the
overall energy expenditures
includes the development of infrastructures and standards needed for
synergistic control and actuation mechanisms at the system level
◦ From chip-level systems such as processors and operating systems
to data centers, including IT equipment hardware, software,
cooling systems
◦ Multi-scale control systems, including scalability and stability
issues for both traditional reactive control mechanisms as well as
proactive control mechanisms based on reduced order models for
predicting workload and thermal trends: the resulting system is
self-sensing and self-regulating.
◦ Mechanisms for utilizing waste heat
14. Five-Year Goals for thermal
management infrastructure
•Determining key intrinsic energy consumption inefficiencies at
every level from devices to entire systems
•Techniques for managing data centers synergistically using
predictive models for computing workload and thermal trends,
driven by live transient models in the control loop
•Airflow management techniques based on fast, compact
models that are continuously validated with live data and using
these models to reduce the overall energy expenditures.
•Techniques for improving the energy efficiency of buildings
and containers.
15. IAB = Industrial
Advisory Board 2011 2012 2013 2014 2015
Holistic Exergy based system modeling Back end waste energy recovery modeling
Multi-scale subsystem Multi-scale system Gaps &
Modeling models (e.g.,aisle) models (e.g., room/bldg) Refinement
Cyber models (compute loads,etc)
Component Subsystem Subsystem System
Validation & Testbed Eval Testbed
Validation Validation
Verification Demos Demos
Hardware & Software Infrastructures
Infrastructures, Integrated
Software Job Scheduling
Decision & Energy Mgmt
Airflow & Water Management
Algorithms
Control Scalable Control Algorithms
Center IAB IAB IAB IAB IAB
Benchmarks Portfolio Benchmark/ Portfolio Benchmark/
Management & Metrics Assessment Assessment Assessment Assessment
16.
17. Typical raised floor data center configuration with alternating hot and cold aisle arrangement.
• Model Details:
▫ CFD software FloTHERM
▫ 20 Racks; 4 CRACs
▫ k-ε turbulence model
▫ Boussinesq approximation
▫ 230k grid cells
▫ Plenum depth: 0.3 m
▫ Tile resistance: 50%
▫ 6 monitor points at rack
inlet. Plan view of the 3D View of the the data center under
rack layout for room
17
study.
18. Team:
Sammakia and Murray (BU),
Agonafer (UTA)
60
Inlet Temperature
Goals:
50
40
Develop verified thermal analysis
30
20
tools to provide real time 10
assessment of Data Center
0 50 100 150 200 250 300 350 400
Time (sec)
operating conditions to enable
Rack A1
operation as dynamic self-
100
% Supply
sensing and self-regulating 60
0 200 400
systems Time (sec)
Enable Data Centers to run at Outcomes:
optimal or near optimal Data showing measured thermal capacity values for
conditions in order to minimize different servers.
energy consumption for a Modeling methodology for data centers transient
broad range of specified response.
performance metrics Design guidelines for thermal management of data
center accounting for both steady state and dynamic
loads.
19. Assessing the Effectiveness of Cold Aisle Containment on the Efficiency of the Data Center Cooling
Infrastructure
Purpose:
◦ Eliminate hot air recirculation.
◦ Eliminate short circuiting of cold air.
◦ Energy efficiency!
Previous problems with fire codes.
Diagram of a data center with hot air recirculation and cold air short
circuiting. (Caption courtesy of www.42U.com) 19
20. Assessing the Effectiveness of Cold Aisle Containment on the Efficiency of the Data Center Cooling
Infrastructure
Cannot totally seal aisle due to pressure and
noise problems.
3 configurations are used for perforated ceiling.
Aim is to keep rack inlet temperature for a cold
aisle uniform.
(a) 3D model in FloTHERM showing perforated ceiling. (b) 3 different configurations used for perforated
ceiling. 20
21. Assessing the Effectiveness of Cold Aisle Containment on the Efficiency of the Data Center Cooling
Infrastructure
All racks powered at 32 kW and CRAC supply is 100%
Rapid drop in inlet
temperature when
sealing the cold
aisle.
Very little variation Maximum rack temperature for various
in inlet temperature perforations.
per rack.
Tile airflow
uniformity and
lower hot air
recirculation
Delta temperature of maximum and minimum
temperature across each rack for various 21
22. Assessing the Effectiveness of Cold Aisle Containment on the Efficiency of the Data Center Cooling
Infrastructure
Average variation of 8 oC with no
containment
Configuration I: average 𝑇 𝑣 is 0.6 oC for 20%
open ceiling.
Temperature profile at height of 1750 mm.
Temperature profile at height of 1750 mm.
22
23. Fan and system curves for CRACs and servers
Steady state and transient models
Impact of thermal capacity
Impact of airflow leaks
Overall thermodynamic models
24. Development of a Compact Model for a Server
Pictures of Top view of the server showing the details of the thermocouple locations (red
thermocouples attached dots). 24
to the server.
25. Development of a Compact Model for a Server
350 12000
Fan 1
Fan Speed (RPM)
300 10000 Fan 2
Power (W)
Fan 3
250
8000 Fan 4
200
6000
150
100 4000
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
Time (s) Time (s)
Measured power. Measured fan speed.
100
Measured Graphic Card
Measured HDD
Measured RAM 1
90 Measured RAM 2
CPU 1 Heatsink Bottom
CPU 1 Heatsink Top
CPU 2 Heatsink Bottom
80 CPU 2 Heatsink Top
Server Graphic Card
Server HDD
Server RAM 1
70 Server RAM 2
Temperature (oC)
Server CPU 1
Server CPU 2
60
50
40
30
20
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Time (s)
Zoomed in plot of: (a) fan speed,
Temperature measurements of different components inside the server. (b) RAM temperature. 25
26. Analyze time constants for variable power
Analyze the impact of air flow and temperature
fluctuations
Analyze the impact of CRAC failures
Collaborate with Georgia Tech and its partners on
infrastructure measurements
Collaborate with UT on developing a simple
control model for airflow
27. Research agenda is unique in its use of
techniques that synergistically combine: (a)
computing (b) innovative cooling solutions and
thermal management techniques and, (c)
adaptive, pro-active and scalable control system
concepts in devising a system-level solution.
Relevant research is already supported by many
sources, including NSF and industry.
Designation as a NSF I/UCRC is formalizing and
expanding current collaborations with industry
and is enabling development of practical,
deployable, high-impact solutions.
28. Bahgat Sammakia
Interim Vice President for Research
Director
The NSF I/UCRC in Energy-Efficient Electronic Systems
Kand Ghose
Chair, Department of Computer Science
Site Director
The NSF I/UCRC in Energy-Efficient Electronic Systems
Binghamton University
Binghamton, NY
13902-6000
http://s3ip.binghamton.edu