Presentation by Jose M. Moya at the IEEE Region 8 SB & GOLD Congress (25 – 29 July, 2012).
The current techniques for data center energy optimization, based on
efficiency metrics like PUE, pPUE, ERE, DCcE, etc., do not take into
account the static and dynamic characteristics of the applications and
resources (computing and cooling). However, the knowledge about the
current state of the data center, the past history, the resource
characteristics, and the characteristics of the jobs to be executed
can be used very effectively to guide decision-making at all levels in
the datacenter in order to minimize energy needs. For example, the
allocation of jobs on the available machines, if done taking into
account the most appropriate architecture for each job from the
energetic point of view, and taking into account the type of jobs that
will come later, can reduce energy needs by 30%.
Moreover, to achieve significant reductions in energy consumption of
state-of-the-art data centers (low PUE) is becoming increasingly
important a comprehensive and multi-level approach, ie, acting on
different abstraction levels (scheduling and resource allocation,
application, operating system, compilers and virtual machines,
architecture, and technology), and at different scopes (chip, server,
rack, room, and multi-room).
Exploring the Future Potential of AI-Enabled Smartphone Processors
Energy-efficient data centers: Exploiting knowledge about application and resources
1. CAMPUS OF
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
Energy-efficient data centers:
Exploiting knowledge about
application and resources
José M. Moya <jm.moya@upm.es>
Integrated Systems Laboratory
José M.Moya | Madrid (Spain), July 27, 2012 1
2. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Data centers
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 2
3. CAMPUS OF
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 3
4. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Power distribution
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 4
5. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Power distribution (Tier 4)
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 5
6. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Contents
“Ingeniamos el futuro”
• Motivation
• Our approach
– Scheduling and resource
management
– Virtual machine
optimizations
– Centralized management
of low-power modes
– Processor design
• Conclusions
José M.Moya | Madrid (Spain), July 27, 2012 6
7. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Motivation
“Ingeniamos el futuro”
• Energy consumption of data centers
– 1.3% of worldwide energy production in 2010
– USA: 80 mill MWh/year in 2011 = 1,5 x NYC
– 1 data center = 25 000 houses
• More than 43 Million Tons of CO2 emissions per
year (2% worldwide)
• More water consumption than many industries
(paper, automotive, petrol, wood, or plastic)
Jonathan Koomey. 2011. Growth in Data center electricity use 2005 to 2010
José M.Moya | Madrid (Spain), July 27, 2012 7
8. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Motivation
“Ingeniamos el futuro” 35000
World server installed base
30000
• It is expected for total data 25000
(thousands)
20000 High-end servers
center electricity use to 15000 Mid-range servers
10000
exceed 400 GWh/year by 5000
Volume servers
2015. 0
2000 2005 2010
• The required energy for 5,75 Million new servers per year
cooling will continue to be at 10% unused servers (CO2 emissions
least as important as the similar to 6,5 million cars)
energy required for the
300
computation. 250 Infrastructure
(billion kWh/year)
Electricity use
200 Communications
• Energy optimization of future 150 Storage
data centers will require a 100 High-end servers
50 Mid-range servers
global and multi-disciplinary 0 Volume servers
approach. 2000 2005 2010
José M.Moya | Madrid (Spain), July 27, 2012 8
9. CAMPUS OF Temperature-dependent
INTERNATIONAL
EXCELLENCE reliability problems
“Ingeniamos el futuro”
✔
Electromigration (EM)
✖
Time-dependent
dielectric-
breakdown (TDDB)
Stress
migration (SM)
✖
✖ Thermal
cycling (TC)
José M.Moya | Madrid (Spain), July 27, 2012 9
10. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Cooling a data center
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 10
11. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Server improvements
“Ingeniamos el futuro”
• Virtualization
- 27%
• Energy Star server
conformance
= 6.500
• Better capacity
planning 2.500
José M.Moya | Madrid (Spain), July 27, 2012 11
12. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Cooling improvements
“Ingeniamos el futuro”
• Improvements in air flow management and
wider temperature ranges
Energy savings
up to 25%
25.000
Return of investment
in only 2 years
José M.Moya | Madrid (Spain), July 27, 2012 12
13. CAMPUS OF
INTERNATIONAL Infrastructure improvements
EXCELLENCE
“Ingeniamos el futuro”
AC DC
– 20% reduction of power losses in the
conversion process
– 47 million dollars savings of real-state costs
– Up to 97% efficiency, energy saving enough to
power an iPad during 70 million years
José M.Moya | Madrid (Spain), July 27, 2012 13
14. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Best practices
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 14
15. CAMPUS OF
And…
what about IT people?
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 15
16. CAMPUS OF
PUE
Power Usage Effectiveness
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
• State of the Art: PUE ≈ 1,2
– The important part is IT energy consumption
– Current work in energy efficient data centers is
focused in decreasing PUE
– Decreasing PIT does not decrease PUE, but it is seen in
the electricity bill
• But how can we reduce PIT ?
José M.Moya | Madrid (Spain), July 27, 2012 16
17. CAMPUS OF
Potential energy savings
by abstraction level
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 17
18. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Our approach
“Ingeniamos el futuro”
• Global strategy to allow the use of multiple
information sources to coordinate decisions in order
to reduce the total energy consumption
• Use of knowledge about the energy demand
characteristics of the applications, and
characteristics of computing and cooling resources
to implement proactive optimization techniques
José M.Moya | Madrid (Spain), July 27, 2012 18
19. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Holistic approach
“Ingeniamos el futuro”
Chip Server Rack Room Multi-
room
Sched & alloc 2 1
app
OS/middleware
Compiler/VM 3 3
architecture 4 4
technology 5
José M.Moya | Madrid (Spain), July 27, 2012 19
20. CAMPUS OF
1. Room-level resource
INTERNATIONAL
EXCELLENCE management
“Ingeniamos el futuro”
Chip Server Rack Room Multi-
room
Sched & alloc 2 1
app
OS/middleware
Compiler/VM 3 3
architecture 4 4
technology 5
José M.Moya | Madrid (Spain), July 27, 2012 20
21. CAMPUS OF
INTERNATIONAL Leveraging heterogeneity
CCGrid 2012
EXCELLENCE
“Ingeniamos el futuro”
• Use heterogeneity to minimize energy
consumption from a static/dynamic point of view
– Static: Finding the best data center set-up, given a
number of heterogeneous machines
– Dynamic: Optimization of task allocation in the
Resource Manager
• We show that the best solution implies an
heterogeneous data center
– Most data centers are heterogeneous (several
generations of computers)
M. Zapater, J.M. Moya, J.L. Ayala. Leveraging Heterogeneity for
Energy Minimization in Data Centers, CCGrid 2012
José M.Moya | Madrid (Spain), July 27, 2012 21
22. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Current scenario
“Ingeniamos el futuro”
Scheduler Resource
WORKLOAD
Manager
Execution
José M.Moya | Madrid (Spain), July 27, 2012 22
23. CAMPUS OF
Potential improvements
with best practices
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
Total power (computing and cooling) for various scheduling approaches
1400 max computing power, worst thermal placement
min computing power, worst thermal placemenit
optimal computing+cooling
1200 optimal computing+cooling, shut off idles
optimal computing+cooling, shut off idles, no recirculation
1000
Power (KW)
savings by minimizing computing power
savings by minimizing the recirculation’s effect
800 savings by turning off idle machines
unaddressed heat recirculation cost
600 basic (unavoidable) cost
400
200
0
0 20 40 60 80 100
job size relative to data center capacity (%)
José
operation cost (in kilowatts) for various “savings
Fig. 3. Data center M.Moya | Madrid (Spain), July 27, 2012 23
24. energy consume
energy consume
20
Cooling-aware scheduling and
100
15
CAMPUS OF
resource allocation
10
INTERNATIONAL
50
EXCELLENCE 5
0
iMPACT Lab (Arizona State U)
0
“Ingeniamos el futuro” FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT
(a) (b)
Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers on Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers off
40 Energy consumption, Scenario (a) 40 jobs, 25014 core-hours, idle servers off
Energy consumption, Scenario (a) 40 jobs, 25014 core-hours,energy
cooling idle servers on cooling energy
computing energy computing energy
40 cooling energy
300 Throughput 0.580 jobs/hr 0.580 jobs/hr 0.349 jobs/hr 0.580 cooling energy
jobs/hr 0.254 jobs/hr 35 computing energy
200 Turnaround time 8.98 hr computing energy
8.98 hr 12.17 hr 8.98 hr 48.49 hr
Throughput 0.580 jobs/hr 0.580 jobs/hr 0.349 jobs/hr 0.580 jobs/hr 0.427 jobs/hr
Alg. runtime 170 ms 186 ms 397 ms 40.8 min 88.6 min 35
250 Energy savings 0.197 jobs/hr
Throughput 0% 0.197 jobs/hr
1.7% 0.172 jobs/hr
4.1% 0.197 jobs/hr
3.6% 0.163 jobs/hr
4.7% 30 Turnaround time
Throughput
8.98 hr
0.197 jobs/hr
8.98 hr
0.197 jobs/hr
12.17 hr
0.172 jobs/hr
8.98 hr
0.197 jobs/hr
17.75 hr
0.163 jobs/hr
Alg. runtime 171 ms 186 ms 397 ms 42 min 100 min
energy consumed (GJ) (GJ)
energy consumed (GJ) (GJ)
Turnaround time 18.41 hr 18.41 hr 20.75 hr 18.41 hr 51.75 hr Turnaround time 18.41 hr 18.41 hr 20.75 hr 18.41 hr 38.02 hr
Alg. runtime 3.4 ms 6.9 ms 213 ms 23 min 40 min 30 Energy savings 0% 4.0% 14.6% 14.2% 15.1%
25 Alg. runtime 3.4 ms 6.9 ms 213 ms 23 min 43 min
energy consumed
energy consumed
150
200
Energy savings 0% 6.2% 8.6% 8.7% 10.2%
Energy savings 0% 11.8% 54.7% 21.8% 60.5%
25
20
150
20
100 15
100 15
10
50
50 10
5
5
0 0
FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT
0 0
FCFS-FF (c)
FCFS-LRH EDF-LRH FCFS-Xint SCINT FCFS-FF (d)
FCFS-LRH EDF-LRH FCFS-Xint SCINT
Energy consumption, Scenario (c)(a) jobs, 45817 core-hours, idle servers on
174 Energy consumption, Scenario (c)(b) jobs, 45817 core-hours, idle servers off
174
Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers on
cooling energy Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers off
cooling energy
computing energy computing energy
450
cooling energy 40 cooling energy
Throughput 0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr energy
computing 0.561 jobs/hr 100 Throughput computing
0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr energy
0.590 jobs/hr
400 Turnaround time 9.99 hr 9.99 hr 13.39 hr 9.99 hr 61.49 hr
300 Turnaround time 9.99 hrjobs/hr 9.99 hrjobs/hr 13.39 hr
Throughput 0.580 0.580
9.99 hr 65.38 hr
0.349 jobs/hr 0.580 jobs/hr 0.254 jobs/hr 35 Alg. runtime 173 ms 191 ms 346 ms 21 min 147 min
Alg. runtime time 173 ms
Turnaround 8.98 hr 196 ms
8.98 hr 346 ms
12.17 hr 20 min
8.98 hr 142 min
48.49 hr Energy savings 0.0% 7.5% 17.3% 25.7% 41.4%
350 Throughput 0.580 jobs/hr 0.580 jobs/hr 0.349 jobs/hr 0.580 jobs/hr 0.427 jobs/hr
energy consumed (GJ) (GJ)
energy consumed (GJ) (GJ)
Energy savings 170 ms
Alg. runtime 0% 2.5%
186 ms 5.9%
397 ms 9.4%
40.8 min 12.5%
88.6 min 80
250 Energy savings 0% 1.7% 4.1% 3.6% 4.7% 30 Turnaround time 8.98 hr 8.98 hr 12.17 hr 8.98 hr 17.75 hr
300 Alg. runtime 171 ms 186 ms 397 ms 42 min 100 min
energy consumed
energy consumed Energy savings 0% 4.0% 14.6% 14.2% 15.1%
25
250
200 60
200 20
150 40
150
15
100
100
20
10
50
50
5
0 0
FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT
0 0
FCFS-FF (e)
FCFS-LRH EDF-LRH FCFS-Xint SCINT FCFS-FF (f)
FCFS-LRH EDF-LRH FCFS-Xint SCINT
(c) (d)
Fig. 8. Energy comparison of the simulated schemes for the three scenarios. The plots correspond in respective positions to the plots of Figure 7.
Energy consumption, Scenario (c) 174 jobs, 45817 core-hours, idle servers on Energy consumption, Scenario (c) 174 jobs, 45817 core-hours, idle servers off
José M.Moya | cooling energy (Spain), July 27, 2012
Madrid 24 cooling energy
policy used in the data center, which enables energy execution as soon as they arrive if the queue is empty and the data
450
computing
job computing energy
Throughput 0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr 0.561 jobs/hr 100 Throughput 0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr 0.590 jobs/hr
400
25. CAMPUS OF Application-aware scheduling and
INTERNATIONAL
EXCELLENCE resource allocation
LSI-UPM
“Ingeniamos el futuro”
Resource
WORKLOAD Manager
(SLURM)
Execution
Profiling and Energy
Classification Optimization
José M.Moya | Madrid (Spain), July 27, 2012 25
26. CAMPUS OF Application-aware scheduling and
INTERNATIONAL
EXCELLENCE resource allocation
Scenario
“Ingeniamos el futuro”
• Workload:
– 12 tasks from SPEC CPU INT 2006
– Random workload composed by 2000 tasks, divided into
job sets
– Random job set arrival time
• Servers:
José M.Moya | Madrid (Spain), July 27, 2012 26
27. CAMPUS OF Application-aware scheduling and
INTERNATIONAL
EXCELLENCE resource allocation
Energy profiling
“Ingeniamos el futuro”
Resource
WORKLOAD Manager
(SLURM)
Execution
Profiling and Energy
Classification Optimization
Energy profiling
José M.Moya | Madrid (Spain), July 27, 2012 27
28. CAMPUS OF
INTERNATIONAL Workload characterization
EXCELLENCE
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 28
29. CAMPUS OF Application-aware scheduling and
INTERNATIONAL
EXCELLENCE resource allocation
“Ingeniamos el futuro”
Optimization
Resource
WORKLOAD Manager
(SLURM)
Execution
Profiling and Energy
Classification Optimization
Energy Minimization:
• Minimization subjected to constraints
• MILP problem (solved with CPLEX)
• Static and Dynamic
José M.Moya | Madrid (Spain), July 27, 2012 29
30. CAMPUS OF Application-aware scheduling and
INTERNATIONAL
EXCELLENCE resource allocation
“Ingeniamos el futuro”
Static optimization
• Definition of optimal data center
– Given a pool of 100 servers of each kind
– 1 job set from workload
– The optimizer chooses the best selection of servers
– Constraints of cost and space
Best solution:
• 40 Sparc
• 27 AMD
Savings:
• 5 a 22% energy
• 30% time
José M.Moya | Madrid (Spain), July 27, 2012 30
31. CAMPUS OF Application-aware scheduling and
INTERNATIONAL
EXCELLENCE resource allocation
“Ingeniamos el futuro”
Dynamic optimization
• Optimal workload allocation
– Complete workload (2000 tasks)
– Good enough resource allocation in terms of energy (not
the best)
– Run-time evaluation and optimization
Energy savings
ranging from 24%
to 47%
José M.Moya | Madrid (Spain), July 27, 2012 31
32. CAMPUS OF Application-aware scheduling and
INTERNATIONAL
EXCELLENCE resource allocation
“Ingeniamos el futuro”
Conclusions
• First proof-of-concept regarding the use of
heterogeneity to save energy
• Automatic solution
• Automatic processor selection offers notable energy
savings
• Easy implementation in real scenarios
– SLURM Resource Manager
– Realistic workloads and servers
José M.Moya | Madrid (Spain), July 27, 2012 32
33. CAMPUS OF
2. Server-level resource
INTERNATIONAL
EXCELLENCE management
“Ingeniamos el futuro”
Chip Server Rack Room Multi-
room
Sched & alloc 2 1
app
OS/middleware
Compiler/VM 3 3
architecture 4 4
technology 5
José M.Moya | Madrid (Spain), July 27, 2012 33
34. CAMPUS OF Scheduling and resource allocation
INTERNATIONAL
EXCELLENCE policies in MPSoCs
“Ingeniamos el futuro”
UCSD – System Energy Efficiency Lab
A. Coskun , T. Rosing , K. Whisnant and K. Gross "Static and dynamic temperature-
aware scheduling for multiprocessor SoCs", IEEE Trans. Very Large Scale Integr. Syst.,
vol. 16, no. 9, pp.1127 -1140 2008
Fig. 3. Distribution of thermal hot spots, with with DPM (ILP).
Fig. 3. Distribution of thermal hot spots, DPM (ILP). Fig. 4. Distribution of spatial gradients, with with DPM (ILP).
Fig. 4. Distribution of spatial gradients, DPM (ILP).
A. Static Scheduling Techniques
A. Static Scheduling Techniques hot spots. While Min-Th reduces the spatial differentials
hot spots. While Min-Th reduces the highhigh spatial differentials
We We next provideextensive comparison of the ILP ILP based above 15 we observe a substantial increase in the spatial
next provide an an extensive comparison of the based above 15 C, C, we observe a substantial increase in the spatial
José M.Moya | Min-Th&Sp. gradients
techniques. We refer to to static approach as as Madrid (Spain), July 27, 2012 above C. C. In contrast, method achieves lower
techniques. We referour our static approach Min-Th&Sp. gradients above 10 10 In contrast,34 our method achieves lower
our
As discussedSection III, we implemented the ILP ILP min- and and more balanced temperature distribution in die. die.
As discussed in in Section III, we implemented the for for min- more balanced temperature distribution in the the
35. CAMPUS OF Scheduling and resource allocation
INTERNATIONAL
EXCELLENCE policies in MPSoCs
“Ingeniamos el futuro”
• Energy characterization of applications allows
to define proactive scheduling and resource
allocation policies, minimizing hotspots
• Hotspot reduction allows to raise cooling
temperature
+1oC means around 7% cooling energy savings
José M.Moya | Madrid (Spain), July 27, 2012 35
36. CAMPUS OF
3. Application-aware and
INTERNATIONAL
EXCELLENCE resource-aware virtual machine
“Ingeniamos el futuro”
Chip Server Rack Room Multi-
room
Sched & alloc 2 1
app
OS/middleware
Compiler/VM 3 3
architecture 4 4
technology 5
José M.Moya | Madrid (Spain), July 27, 2012 36
37. CAMPUS OF
JIT compilation in
virtual machines
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
• Virtual machines compile
(JIT compilation) the
applications into native
code for performance
reasons
• The optimizer is general-
purpose and focused in
performance
optimization
José M.Moya | Madrid (Spain), July 27, 2012 37
38. CAMPUS OF
JIT compilation for
energy minimization
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
Back-end
Front-end Code generator
Optimizer
• Application-aware compiler
– Energy characterization of applications and
transformations
– Application-dependent optimizer
– Global view of the data center workload
• Energy optimizer
– Currently, compilers for high-end processors oriented
to performance optimization
José M.Moya | Madrid (Spain), July 27, 2012 38
39. CAMPUS OF
Energy saving potential for
the compiler (MPSoCs)
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
T. Simunic, G. de Micheli, L. Benini, and M. Hans. “Source code optimization and
profiling of energy consumption in embedded systems,” International Symposium on
System Synthesis, pages 193 – 199, Sept. 2000
– 77% energy reduction in MP3 decoder
FEI, Y., RAVI, S., RAGHUNATHAN, A., AND JHA, N. K. 2004. Energy-optimizing source
code transformations for OS-driven embedded software. In Proceedings of the
International Conference VLSI Design. 261–266.
– Up to 37,9% (mean 23,8%) energy savings in
multiprocess applications running on Linux
José M.Moya | Madrid (Spain), July 27, 2012 39
40. CAMPUS OF
4. Global automatic
INTERNATIONAL
EXCELLENCE
management of low-power
“Ingeniamos el futuro”
modes
Chip Server Rack Room Multi-
room
Sched & alloc 2 1
app
OS/middleware
Compiler/VM 3 3
architecture 4 4
technology 5
José M.Moya | Madrid (Spain), July 27, 2012 40
41. CAMPUS OF
DVFS – Dynamic Voltage
and Frequency Scaling
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
• As supply voltage decreases, power decreases
quadratically
• But delay increases (performance decreases)
only linearly
• The maximum frequency also decreases
linearly
• Currently, low-power modes, if used, are
activated by inactivity of the server operating
system
José M.Moya | Madrid (Spain), July 27, 2012 41
42. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Room-level DVFS
“Ingeniamos el futuro”
• To minimize energy consumption, changes
between modes should be minimized
• There exist optimal algorithms for a known
task set (YDS)
• Workload knowledge allows to globally
schedule low-power modes without any
impact in performance
José M.Moya | Madrid (Spain), July 27, 2012 42
43. CAMPUS OF
INTERNATIONAL Parallelism to save energy
EXCELLENCE
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 43
44. CAMPUS OF 5. Temperature-aware floorplanning of
INTERNATIONAL
EXCELLENCE MPSoCs and many-cores
“Ingeniamos el futuro”
Chip Server Rack Room Multi-
room
Sched & alloc 2 1
app
OS/middleware
Compiler/VM 3
architecture 4 4
technology 5
José M.Moya | Madrid (Spain), July 27, 2012 44
45. CAMPUS OF
Temperature-aware
floorplanning
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
José M.Moya | Madrid (Spain), July 27, 2012 45
46. Average MaxTemp reduction: 12 oC
Potential energy savings
CAMPUS OF
Larger temperature reductions for benchmarks
with floorplanning
INTERNATIONAL
with higher maximum temperature
EXCELLENCE
“Ingeniamos el futuro”
For many benchmarks, temperature reducions are
Y. Han, I. Koren, and C. A. Moritz. Temperature Aware Floorplanning. In Proc. of the
larger than 20 oC
Second Workshop on Temperature-Aware Computer Systems, June 2005
Maximum Temperature original modified
140
120
100
80
60
40
20
0
wupwise
twolf
swim
gzip
mgrid
mcf
lucas
applu
ammp
bzip2
crafty
fma3d
perlbmk
vortex
avg
apsi
vpr
equake
facerec
gcc
mesa
eon
gap
art
parser
– Up to 21oC reduction of maximum temperature
– Mean: -12oC in maximum temperature
– Better results in the most critical examples
José M.Moya | Madrid (Spain), July 27, 2012 46
47. CAMPUS OF
Temperature-aware
INTERNATIONAL
EXCELLENCE floorplanning in 3D chips
“Ingeniamos el futuro”
• 3D chips are getting interest due to:
– Scalability: reduces 2D equivalent
area
– Performance: shorter wire length
– Reliability: less wiring
• Drawback:
– Huge increment of hotspots
compared with 2D equivalent designs
José M.Moya | Madrid (Spain), July 27, 2012 47
48. CAMPUS OF
Temperature-aware
floorplanning in 3D chips
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
• Up to 30oC reduction per layer in a 3D chip
with 4 layers and 48 cores
José M.Moya | Madrid (Spain), July 27, 2012 48
49. CAMPUS OF
There is still much more
to be done
INTERNATIONAL
EXCELLENCE
“Ingeniamos el futuro”
• Smart Grids
– Consume energy when everybody else does not
– Decrease energy consumption when everybody
else is consuming
• Reducing the electricity bill
– Variable electricity rates
– Reactive power coefficient
– Peak energy demand
José M.Moya | Madrid (Spain), July 27, 2012 49
50. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Conclusions
“Ingeniamos el futuro”
• Reducing PUE is not the same as reducing energy
consumption
– IT energy consumption dominates in state-of-the-art data
centers
• Application and resources knowledge can be effectively
used to define proactive policies to reduce the total energy
consumption
– At different levels
– In different scopes
– Taking into account cooling and computation at the same time
• Proper management of the knowledge of the data center
thermal behavior can reduce reliability issues
• Reducing energy consumption is not the same as reducing
the electricity bill
José M.Moya | Madrid (Spain), July 27, 2012 50
51. CAMPUS OF
INTERNATIONAL
EXCELLENCE
Contact
“Ingeniamos el futuro”
José M. Moya
+34 607 082 892
jm.moya@upm.es
ETSI de Telecomunicación, B104
Avenida Complutense, 30
Madrid 28040, Spain
Gracias:
José M.Moya | Madrid (Spain), July 27, 2012 51