Share
Energy-efficient data centers: Exploiting knowledge about application and resources
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).
GreenLSI
Transcript
51 slides
-
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Energy-efficient data centers:
Exploiting knowledge about application and resources José M. Moya <[email protected]> Integrated Systems Laboratory José M.Moya | Madrid (Spain), July 27, 2012 1 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Data centers José
M.Moya | Madrid (Spain), July 27, 2012 2 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE José M.Moya |
Madrid (Spain), July 27, 2012 3 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Power distribution José
M.Moya | Madrid (Spain), July 27, 2012 4 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Power distribution (Tier
4) José M.Moya | Madrid (Spain), July 27, 2012 5 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Contents • MoHvaHon
• Our approach – Scheduling and resource management – Virtual machine opHmizaHons – Centralized management of low-‐power modes – Processor design • Conclusions José M.Moya | Madrid (Spain), July 27, 2012 6 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Motivation José M.Moya
| Madrid (Spain), July 27, 2012 7 • Energy consumpHon of data centers – 1.3% of worldwide energy producHon 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 consumpHon than many industries (paper, automoHve, petrol, wood, or plasHc) Jonathan Koomey. 2011. Growth in Data center electricity use 2005 to 2010 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Motivation José M.Moya
| Madrid (Spain), July 27, 2012 8 • It is expected for total data center electricity use to exceed 400 GWh/year by 2015. • The required energy for cooling will conHnue to be at least as important as the energy required for the computaHon. • Energy opHmizaHon of future data centers will require a global and mulH-‐disciplinary approach. 0 5000 10000 15000 20000 25000 30000 35000 2000 2005 2010 World server installed base (thousands) High-‐end servers Mid-‐range servers Volume servers 0 50 100 150 200 250 300 2000 2005 2010 Electricity use (billion kWh/year) Infrastructure CommunicaHons Storage High-‐end servers Mid-‐range servers Volume servers 5,75 Million new servers per year 10% unused servers (CO2 emissions similar to 6,5 million cars) -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Temperature-dependent reliability problems
José M.Moya | Madrid (Spain), July 27, 2012 9 Time-‐dependent dielectric-‐ breakdown (TDDB) ElectromigraHon (EM) Stress migraHon (SM) Thermal cycling (TC) ✔ ✖ ✖ ✖ -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Cooling a data
center José M.Moya | Madrid (Spain), July 27, 2012 10 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE • VirtualizaHon
-‐ 27% • Energy Star server conformance = 6.500 • Bejer capacity planning 2.500 Server improvements José M.Moya | Madrid (Spain), July 27, 2012 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . Servidores 5,75 millones de nuevos servido- res se instalan cada año para mantenerse al ritmo de creci- miento de los servicios on-line, y todavía aproximadamente el 10% de los servidores instalados no se utilizan debido a sobrestima- ciones conservadoras a la hora de planificar las necesidades de almacenamiento. La energía utilizada para los servidores en desuso podría compensar las emisiones de 6,5 millones de coches. de los servidores. En un centro de datos convencional, algunas de estas mejoras pueden reducir su impacto en las siguientes cifras: (excluyendo el software) $ $ $ $ A menudo los servidores se sobredimensionan para afrontar picos de demanda, lo que significa que como media suelen funcionar sólo al 20% 2.500 6,5 millones = 6.500 El equivalente a retirar 6.500 coches de las carreteras, me- diante la utilización de servidores acordes a Energy Star, lo que re- duciría el consumo eléctrico de los centros de datos en 82.000 megavatios-hora. 10% no se utilizan US 27% Reducir un 27% el consumo energético mediante la virtuali- zación, lo que reduce la capaci- dad productiva no empleada. El equivalente a la energía consumida por 2.500 hogares en EEUU, mediante una mejor planifica- ción de la capacidad. -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Cooling improvements •
Improvements in air flow management and wider temperature ranges José M.Moya | Madrid (Spain), July 27, 2012 12 Energy savings up to 25% 25.000 Return of investment in only 2 years -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE AC è DC
– 20% reducHon 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 Infrastructure improvements José M.Moya | Madrid (Spain), July 27, 2012 13 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE
Best practices José M.Moya | Madrid (Spain), July 27, 2012 14 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE And… what about
IT people? José M.Moya | Madrid (Spain), July 27, 2012 15 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE PUE Power Usage
Effectiveness • State of the Art: PUE ≈ 1,2 – The important part is IT energy consumpHon – 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 !"# = 1 !"#$ = !!"!#$ !!" =! ! !!!!!!!!!!= !!"#$% + !!""#$%& + !!"# !!"#$% ≈ !!""#$%& + !!" !!" ! -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Potential energy savings
by abstraction level José M.Moya | Madrid (Spain), July 27, 2012 17 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Our approach •
Global strategy to allow the use of mulHple informaHon sources to coordinate decisions in order to reduce the total energy consumpHon • Use of knowledge about the energy demand characterisHcs of the applicaEons, and characterisHcs of compuEng and cooling resources to implement proacEve opEmizaEon techniques José M.Moya | Madrid (Spain), July 27, 2012 18 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Holistic approach José
M.Moya | Madrid (Spain), July 27, 2012 19 Chip Server Rack Room MulE-‐ room Sched & alloc 2 1 app OS/middleware Compiler/VM 3 3 architecture 4 4 technology 5 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE 1. Room-level resource
management José M.Moya | Madrid (Spain), July 27, 2012 20 Chip Server Rack Room MulE-‐ room Sched & alloc 2 1 app OS/middleware Compiler/VM 3 3 architecture 4 4 technology 5 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Leveraging heterogeneity •
Use heterogeneity to minimize energy consumpHon from a staHc/dynamic point of view – StaEc: Finding the best data center set-‐up, given a number of heterogeneous machines – Dynamic: OpHmizaHon of task allocaHon in the Resource Manager • We show that the best soluHon implies an heterogeneous data center – Most data centers are heterogeneous (several generaHons of computers) 21 CCGrid 2012 José M.Moya | Madrid (Spain), July 27, 2012 M. Zapater, J.M. Moya, J.L. Ayala. Leveraging Heterogeneity for Energy MinimizaHon in Data Centers, CCGrid 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Current scenario 22
WORKLOAD Scheduler Resource Manager ExecuEon José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Potential improvements with
best practices José M.Moya | Madrid (Spain), July 27, 2012 23 ot aisle 0 200 400 600 800 1000 1200 1400 0 20 40 60 80 100 Power (KW) job size relative to data center capacity (%) Total power (computing and cooling) for various scheduling approaches savings by minimizing computing power savings by minimizing the recirculation’s effect savings by turning off idle machines unaddressed heat recirculation cost basic (unavoidable) cost max computing power, worst thermal placement min computing power, worst thermal placemenit optimal computing+cooling optimal computing+cooling, shut off idles optimal computing+cooling, shut off idles, no recirculation Fig. 3. Data center operation cost (in kilowatts) for various “savings modes”. Savings are based on heat recirculation data obtained by -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Cooling-aware scheduling and
resource allocation José M.Moya | Madrid (Spain), July 27, 2012 24 0 50 100 150 200 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (a) 40 jobs, 25014 core-hours, idle servers on Throughput Turnaround time Alg. runtime Energy savings 0.197 jobs/hr 0.197 jobs/hr 0.172 jobs/hr 0.197 jobs/hr 0.163 jobs/hr 18.41 hr 18.41 hr 20.75 hr 18.41 hr 51.75 hr 3.4 ms 6.9 ms 213 ms 23 min 40 min 0% 6.2% 8.6% 8.7% 10.2% cooling energy computing energy (a) 0 5 10 15 20 25 30 35 40 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (a) 40 jobs, 25014 core-hours, idle servers off Throughput Turnaround time Alg. runtime Energy savings 0.197 jobs/hr 0.197 jobs/hr 0.172 jobs/hr 0.197 jobs/hr 0.163 jobs/hr 18.41 hr 18.41 hr 20.75 hr 18.41 hr 38.02 hr 3.4 ms 6.9 ms 213 ms 23 min 43 min 0% 11.8% 54.7% 21.8% 60.5% cooling energy computing energy (b) 0 50 100 150 200 250 300 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers on Throughput Turnaround time Alg. runtime Energy savings 0.580 jobs/hr 0.580 jobs/hr 0.349 jobs/hr 0.580 jobs/hr 0.254 jobs/hr 8.98 hr 8.98 hr 12.17 hr 8.98 hr 48.49 hr 170 ms 186 ms 397 ms 40.8 min 88.6 min 0% 1.7% 4.1% 3.6% 4.7% cooling energy computing energy (c) 0 5 10 15 20 25 30 35 40 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers off Throughput Turnaround time Alg. runtime Energy savings 0.580 jobs/hr 0.580 jobs/hr 0.349 jobs/hr 0.580 jobs/hr 0.427 jobs/hr 8.98 hr 8.98 hr 12.17 hr 8.98 hr 17.75 hr 171 ms 186 ms 397 ms 42 min 100 min 0% 4.0% 14.6% 14.2% 15.1% cooling energy computing energy (d) 300 350 400 450 (GJ) Energy consumption, Scenario (c) 174 jobs, 45817 core-hours, idle servers on Throughput Turnaround time Alg. runtime Energy savings 0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr 0.561 jobs/hr 9.99 hr 9.99 hr 13.39 hr 9.99 hr 65.38 hr 173 ms 196 ms 346 ms 20 min 142 min 0% 2.5% 5.9% 9.4% 12.5% cooling energy computing energy 80 100 (GJ) Energy consumption, Scenario (c) 174 jobs, 45817 core-hours, idle servers off Throughput Turnaround time Alg. runtime Energy savings 0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr 0.590 jobs/hr 9.99 hr 9.99 hr 13.39 hr 9.99 hr 61.49 hr 173 ms 191 ms 346 ms 21 min 147 min 0.0% 7.5% 17.3% 25.7% 41.4% cooling energy computing energy 0 50 100 150 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Turnaround time Alg. runtime Energy savings 18.41 hr 18.41 hr 20.75 hr 18.41 hr 51.75 hr 3.4 ms 6.9 ms 213 ms 23 min 40 min 0% 6.2% 8.6% 8.7% 10.2% (a) 0 5 10 15 20 25 30 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Throughput Turnaround time Alg. runtime Energy savings 0.197 jobs/hr 0.197 jobs/hr 0.172 jobs/hr 0.197 jobs/hr 0.163 jobs/hr 18.41 hr 18.41 hr 20.75 hr 18.41 hr 38.02 hr 3.4 ms 6.9 ms 213 ms 23 min 43 min 0% 11.8% 54.7% 21.8% 60.5% (b) 0 50 100 150 200 250 300 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers on Throughput Turnaround time Alg. runtime Energy savings 0.580 jobs/hr 0.580 jobs/hr 0.349 jobs/hr 0.580 jobs/hr 0.254 jobs/hr 8.98 hr 8.98 hr 12.17 hr 8.98 hr 48.49 hr 170 ms 186 ms 397 ms 40.8 min 88.6 min 0% 1.7% 4.1% 3.6% 4.7% cooling energy computing energy (c) 0 5 10 15 20 25 30 35 40 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (b) 120 jobs, 16039 core-hours, idle servers off Throughput Turnaround time Alg. runtime Energy savings 0.580 jobs/hr 0.580 jobs/hr 0.349 jobs/hr 0.580 jobs/hr 0.427 jobs/hr 8.98 hr 8.98 hr 12.17 hr 8.98 hr 17.75 hr 171 ms 186 ms 397 ms 42 min 100 min 0% 4.0% 14.6% 14.2% 15.1% cooling energy computing energy (d) 0 50 100 150 200 250 300 350 400 450 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (c) 174 jobs, 45817 core-hours, idle servers on Throughput Turnaround time Alg. runtime Energy savings 0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr 0.561 jobs/hr 9.99 hr 9.99 hr 13.39 hr 9.99 hr 65.38 hr 173 ms 196 ms 346 ms 20 min 142 min 0% 2.5% 5.9% 9.4% 12.5% cooling energy computing energy (e) 0 20 40 60 80 100 FCFS-FF FCFS-LRH EDF-LRH FCFS-Xint SCINT energy consumed (GJ) Energy consumption, Scenario (c) 174 jobs, 45817 core-hours, idle servers off Throughput Turnaround time Alg. runtime Energy savings 0.892 jobs/hr 0.892 jobs/hr 0.861 jobs/hr 0.892 jobs/hr 0.590 jobs/hr 9.99 hr 9.99 hr 13.39 hr 9.99 hr 61.49 hr 173 ms 191 ms 346 ms 21 min 147 min 0.0% 7.5% 17.3% 25.7% 41.4% cooling energy computing energy (f) Fig. 8. Energy comparison of the simulated schemes for the three scenarios. The plots correspond in respective positions to the plots of Figure 7. policy used in the data center, which enables job execution as soon as they arrive if the queue is empty and the data center is lightly loaded. In the “idle-on” case (Figure 8a), the total energy consumption using SCINT, EDF-LRH, iMPACT Lab (Arizona State U) -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Application-aware scheduling and
resource allocation 25 LSI-UPM WORKLOAD Resource Manager (SLURM) ExecuEon Profiling and ClassificaEon Energy OpEmizaEon José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Application-aware scheduling and
resource allocation • Workload: – 12 tasks from SPEC CPU INT 2006 – Random workload composed by 2000 tasks, divided into job sets – Random job set arrival Hme • Servers: 26 Scenario José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Application-aware scheduling and
resource allocation 27 Energy profiling WORKLOAD Resource Manager (SLURM) ExecuEon Profiling and ClassificaEon Energy OpEmizaEon Energy profiling José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Workload characterization José
M.Moya | Madrid (Spain), July 27, 2012 28 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Application-aware scheduling and
resource allocation 29 Optimization WORKLOAD Resource Manager (SLURM) ExecuEon Profiling and ClassificaEon Energy OpEmizaEon Energy MinimizaEon: • MinimizaEon subjected to constraints • MILP problem (solved with CPLEX) • StaEc and Dynamic José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Application-aware scheduling and
resource allocation 30 Static optimization • DefiniHon of opHmal data center – Given a pool of 100 servers of each kind – 1 job set from workload – The opHmizer chooses the best selecHon of servers – Constraints of cost and space Best soluEon: • 40 Sparc • 27 AMD Savings: • 5 a 22% energy • 30% Eme José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Application-aware scheduling and
resource allocation 31 Dynamic optimization • OpHmal workload allocaHon – Complete workload (2000 tasks) – Good enough resource allocaHon in terms of energy (not the best) – Run-‐Hme evaluaHon and opHmizaHon Energy savings ranging from 24% to 47% José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Application-aware scheduling and
resource allocation • First proof-‐of-‐concept regarding the use of heterogeneity to save energy • AutomaHc soluHon • AutomaHc processor selecHon offers notable energy savings • Easy implementaHon in real scenarios – SLURM Resource Manager – RealisHc workloads and servers 32 Conclusions José M.Moya | Madrid (Spain), July 27, 2012 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE 2. Server-level resource
management José M.Moya | Madrid (Spain), July 27, 2012 33 Chip Server Rack Room MulE-‐ room Sched & alloc 2 1 app OS/middleware Compiler/VM 3 3 architecture 4 4 technology 5 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Scheduling and resource
allocation policies in MPSoCs A. Coskun , T. Rosing , K. Whisnant and K. Gross "StaHc and dynamic temperature-‐ aware scheduling for mulHprocessor SoCs", IEEE Trans. Very Large Scale Integr. Syst., vol. 16, no. 9, pp.1127 -‐1140 2008 José M.Moya | Madrid (Spain), July 27, 2012 34 Fig. 3. Distribution of thermal hot spots, with DPM (ILP). A. Static Scheduling Techniques We next provide an extensive comparison of the ILP based techniques. We refer to our static approach as Min-Th&Sp. As discussed in Section III, we implemented the ILP for min- imizing thermal hot spots (Min-Th), energy balancing (Bal- En), and energy minimization (Min-En) to compare against Fig. 4. Distribution of spatial gradients, with DPM (ILP). hot spots. While Min-Th reduces the high spatial differentials above 15 C, we observe a substantial increase in the spatial gradients above 10 C. In contrast, our method achieves lower and more balanced temperature distribution in the die. In Fig. 5, we show how the magnitudes of thermal cycles vary with the scheduling method. We demonstrate the average per- Fig. 3. Distribution of thermal hot spots, with DPM (ILP). A. Static Scheduling Techniques We next provide an extensive comparison of the ILP based techniques. We refer to our static approach as Min-Th&Sp. As discussed in Section III, we implemented the ILP for min- imizing thermal hot spots (Min-Th), energy balancing (Bal- Fig. 4. Distribution of spatial gradients, with DPM (ILP). hot spots. While Min-Th reduces the high spatial differentials above 15 C, we observe a substantial increase in the spatial gradients above 10 C. In contrast, our method achieves lower and more balanced temperature distribution in the die. In Fig. 5, we show how the magnitudes of thermal cycles vary UCSD – System Energy Efficiency Lab -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Scheduling and resource
allocation policies in MPSoCs • Energy characterizaHon of applicaHons allows to define proacHve scheduling and resource allocaHon policies, minimizing hotspots • Hotspot reducHon allows to raise cooling temperature +1oC means around 7% cooling energy savings José M.Moya | Madrid (Spain), July 27, 2012 35 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE 3. Application-aware and
resource-aware virtual machine José M.Moya | Madrid (Spain), July 27, 2012 36 Chip Server Rack Room MulE-‐ room Sched & alloc 2 1 app OS/middleware Compiler/VM 3 3 architecture 4 4 technology 5 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE JIT compilation in
virtual machines • Virtual machines compile (JIT compilaHon) the applicaHons into naHve code for performance reasons • The opHmizer is general-‐ purpose and focused in performance opEmizaEon José M.Moya | Madrid (Spain), July 27, 2012 37 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Back-‐end JIT
compilation for energy minimization • ApplicaHon-‐aware compiler – Energy characterizaHon of applicaHons and transformaHons – ApplicaHon-‐dependent opHmizer – Global view of the data center workload • Energy opHmizer – Currently, compilers for high-‐end processors oriented to performance opHmizaHon José M.Moya | Madrid (Spain), July 27, 2012 38 Front-‐end OpHmizer Code generator -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Energy saving potential
for the compiler (MPSoCs) T. Simunic, G. de Micheli, L. Benini, and M. Hans. “Source code opHmizaHon and profiling of energy consumpHon in embedded systems,” InternaHonal Symposium on System Synthesis, pages 193 – 199, Sept. 2000 – 77% energy reducHon in MP3 decoder FEI, Y., RAVI, S., RAGHUNATHAN, A., AND JHA, N. K. 2004. Energy-‐opHmizing source code transformaHons for OS-‐driven embedded soyware. In Proceedings of the InternaHonal Conference VLSI Design. 261–266. – Up to 37,9% (mean 23,8%) energy savings in mulHprocess applicaHons running on Linux José M.Moya | Madrid (Spain), July 27, 2012 39 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE 4. Global automatic
management of low-power modes José M.Moya | Madrid (Spain), July 27, 2012 40 Chip Server Rack Room MulE-‐ room Sched & alloc 2 1 app OS/middleware Compiler/VM 3 3 architecture 4 4 technology 5 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE DVFS – Dynamic
Voltage and Frequency Scaling • As supply voltage decreases, power decreases quadraHcally • But delay increases (performance decreases) only linearly • The maximum frequency also decreases linearly • Currently, low-‐power modes, if used, are acHvated by inacHvity of the server operaHng system José M.Moya | Madrid (Spain), July 27, 2012 41 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Room-level DVFS •
To minimize energy consumpHon, changes between modes should be minimized • There exist opHmal 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 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Parallelism to save
energy Use of Parallelism Use of Parallelism V dd V dd /2 V dd /2 f f /2 f /2 f max f max /2 f max /2 9-17 Swiss Federal Institute of Technology Computer Engineering and Networks Laboratory José M.Moya | Madrid (Spain), July 27, 2012 43 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE 5. Temperature-aware floorplanning
of MPSoCs and many-cores José M.Moya | Madrid (Spain), July 27, 2012 44 Chip Server Rack Room MulE-‐ room Sched & alloc 2 1 app OS/middleware Compiler/VM 3 architecture 4 4 technology 5 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Temperature-aware floorplanning José
M.Moya | Madrid (Spain), July 27, 2012 45 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Potential energy savings
with floorplanning – Up to 21oC reducHon of maximum temperature – Mean: -‐12oC in maximum temperature – Bejer results in the most criHcal examples José M.Moya | Madrid (Spain), July 27, 2012 46 Temperature Reductions Average MaxTemp reduction: 12 oC Larger temperature reductions for benchmarks with higher maximum temperature For many benchmarks, temperature reducions are larger than 20 oC Maximum Temperature 0 20 40 60 80 100 120 140 ammp applu apsi art bzip2 crafty eon equake facerec fma3d gap gcc gzip lucas mcf mesa mgrid parser perlbmk swim twolf vortex vpr wupwise avg original modified Y. Han, I. Koren, and C. A. Moritz. Temperature Aware Floorplanning. In Proc. of the Second Workshop on Temperature-‐Aware Computer Systems, June 2005 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Temperature-aware floorplanning in
3D chips José M.Moya | Madrid (Spain), July 27, 2012 47 • 3D chips are gezng 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 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Temperature-aware floorplanning in
3D chips José M.Moya | Madrid (Spain), July 27, 2012 48 • Up to 30oC reducHon per layer in a 3D chip with 4 layers and 48 cores -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE There is still
much more to be done • Smart Grids – Consume energy when everybody else does not – Decrease energy consumpHon when everybody else is consuming • Reducing the electricity bill – Variable electricity rates – ReacHve power coefficient – Peak energy demand José M.Moya | Madrid (Spain), July 27, 2012 49 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Conclusions • Reducing
PUE is not the same as reducing energy consumpHon – IT energy consumpHon dominates in state-‐of-‐the-‐art data centers • ApplicaHon and resources knowledge can be effecHvely used to define proacEve policies to reduce the total energy consumpHon – At different levels – In different scopes – Taking into account cooling and computaHon at the same Hme • Proper management of the knowledge of the data center thermal behavior can reduce reliability issues • Reducing energy consumpHon is not the same as reducing the electricity bill José M.Moya | Madrid (Spain), July 27, 2012 50 -
“Ingeniamos el futuro”! CAMPUS OF INTERNATIONAL EXCELLENCE Contact José M.Moya
| Madrid (Spain), July 27, 2012 51 José M. Moya +34 607 082 892 [email protected] ETSI de Telecomunicación, B104 Avenida Complutense, 30 Madrid 28040, Spain Gracias: