Date post: | 12-Jan-2016 |
Category: |
Documents |
Upload: | gordon-snow |
View: | 215 times |
Download: | 0 times |
Leveraging Renewable Energy in Data Centers
Ricardo Bianchini on tour2012
Motivation• Datacenters consume large amounts of energy• Energy cost is not the only problem– Brown sources: coal, natural gas…
• Lots of small and medium datacenters• Use renewable sources for datacenters– Solar panels, wind turbines…– “Green" datacenters
Renewable sources has lower environmental impact
Wind (o
ffshore)
Hydroelectr
ic
Wind (o
nshore)
Biogas
Solar therm
al
Biomass
Solar PV
Geoterhemal
Nuclear
Natural gas
Diesel
Heavy oilCoal
0100200300400500600700800900
1000CO2 generation per source (CO2/kWh)
Wastes lastthousands of years
Motivation• Datacenters consume large amounts of energy• Energy cost is not the only problem– Brown sources: coal, natural gas…
• Lots of small and medium datacenters• Use renewable sources for datacenters– Solar panels, wind turbines…– “Green" datacenters
Renewable approaches for“green” datacenters
• Centralized generation– Utilities install renewable power plants– Green pricing: users pay for a green percentage
• Compensation systems– CO2 offsets– Renewable Electricity Credits (RECs)
• Distributed generation– Connect to close renewable power plants– Self-generation
Why distributed generation?
• Energy independence– Stable costs– Resilient to external failures
• Reduce transmission and conversion losses– From 41% to 30% losses (even <5%)
• Long lifetime– Installations can be reused
• Allow local energy management– Control which energy to use and when
Why solar and wind?
• Medium/high availability• Suitable for small/medium installations– Initial cost/W is lower– No wastes– Easy to install– Easy to maintain
Why solar?
• Solar PV cost scalability is linear– Small installations have similar $/W as large– Better for distributed generation
• Solar availability is higher than windWind Solar
Price of PV energy is decreasing19
9719
9819
9920
0020
0120
0220
0320
0420
0520
0620
0720
0820
0920
1020
11
02468
101214
Grid electricity price(2011 c/kWh)
19801983
19861989
19921995
19982001
20042007
20100
5
10
15
20
25
PV panels cost(2011 $/W)
• PV energy already cheaper than utility in some locations [1]
Problems and challenges
• Require large extensions of land– Bad for large datacenters
• Low/medium efficiency– Efficiency lower than 40%– Increasing every year
• Variability– Batteries: losses, economic and environmental cost– Net metering: losses, limited availability– Smarter management
Parasol (remove)
• Construction– Structure– Solar panels
• Installation– Servers: software
• Lessons learned– Not as easy as it seems– Hard to deal with facilities crowd– Setting it on the roof has extra cost– Easier to just put solar panels in the roof and use a regular room– Flexibility for research purposes is hard– Metering everywhere: temperature, power
Parasol
• We are building a µDatacenter• Powered by– PV panels– Electricity grid– Batteries
• Research framework– Manage solar-powered datacenter– Software to exploit renewable energy– Free cooling
Parasol description• Installed on the roof• Steel structure
– Container to host the IT– 10 PV panels: 3 kW
• Backup energy systems– Batteries: 32 kWh– Power grid
• IT equipment– 2 42U racks– 64 Atom servers (so far)– 2 switches
• Cooling system– Free cooling– Air conditioner– Heater
Structure craning and assembling
Container and solar panels
Electrical and cooling
IT equipment
Green systems
• Handle renewable energy variability– Smart energy management
• GreenSlot [SC‘11]– Schedule batch jobs (SLURM)
• GreenHadoop [EUROSYS’12]– Schedule data-processing jobs (MapReduce)
Green systems approach
• Predict green energy availability– Weather forecast
• Schedule jobs– Maximize green energy use– If green not available, consume cheap brown
• May delay jobs but must meet deadlines• Manage data availability• Send to sleep (S3) idle servers to save energy
J1
GreenSlot behavior
J2
Time
J1
J2
Now
Nod
esPow
er
J1J2
Schedule:
Brown electricity priceJob deadlineScheduling window
J1, J2
J1J3
J4
GreenSlot behavior
J2
Time
J1
J2
J4
J3
Nod
esPow
er
J3J4
Schedule:
Now
J3, J4
Brown electricity priceJob deadlineScheduling window
J1
J4
J3
GreenSlot behavior
J2
Time
J2
J1J3
Nod
esPow
er
J4
J4
Schedule: J4 Weather prediction was wrong
NowBrown electricity priceJob deadlineScheduling window
J1
J4
J5J3
GreenSlot behavior
J2
Time
J2
J1J3 J5
Nod
esPow
er
J4
J5
Schedule:
Now
J5
Brown electricity priceJob deadlineScheduling window
Future directions
• Collect data of the data center– Real workloads– Temperatures
• ???
Leveraging Renewable Energy in Data Centers
Ricardo Bianchini on tour2012