Leveraging Renewable Energy in Data Centers Ricardo Bianchini Department of Computer Science Collaborators: Inigo Goiri, Jordi Guitart (UPC/BSC), Md. Haque, William Katsak, Kien Le, Thu D. Nguyen, and Jordi Torres (UPC/BSC)
Outline DC energy usage and carbon footprint Reducing carbon with renewables Our approach Parasol: our solar micro-data center Software for leveraging solar energy Current and future works Conclusions
Billion KWh/year Billion KWh/year Motivation Data centers = machine rooms to giant warehouses Consume massive amounts of energy (electricity) 90 60 30 0 2000 2005 2010 270 240 210 180 150 120 90 60 30 0 2000 2005 2010 Electricity consumption of US DCs [JK 11] Electricity consumption of WW DCs [JK 11]
MMT/year Motivation Electricity comes mostly from burning fossil fuels 100% 80% 60% 40% Others Renewables Nuclear Natural Gas 120 116 112 108 35th 34th 20% Coal 104 0% US World 100 Nigeria Data Centers Czech Rep. Electricity sources in US & WW [DOE 10] CO 2 of world-wide DCs [Mankoff 08] Can we use renewables to reduce this footprint?
Outline DC energy usage and carbon footprint Reducing carbon with renewables: 4 approaches Our approach Parasol: our solar micro-data center Software for leveraging solar energy Current and future works Conclusions
Greening DCs: Compensation approaches Carbon offsets to compensate for brown energy use Pros: DC operator need not worry about renewable energy Encourages DC operators to conserve energy Cons: Offsets are sometimes purchased far away Offsets are an extra cost for businesses Example: Cap-and-trade in the UK
Greening DCs: Grid-centric approaches Pump renewables into the grid Pros: If the grid is available, power is always available DC operator need not worry about renewable plants Plants can be placed at the best possible locations Cons: Power transmission can lead to 40% energy losses Certain renewable technologies are not scalable Dependence on the power grid or diesel generators Example: Google buys wind power from NextEra
Greening DCs: Co-location approaches Build data center near a renewable plant Pros: Reduced energy losses No dependence on the grid Cons: Location may not be good for DCs Power may have already been committed Example: Microsoft builds DC near hydro plant in OR
Greening DCs: Distributed generation Self-generation: Build your own renewable plant Pros: Reduced energy losses Can operate during grid outages without diesel generators Capital costs can be recovered for lower energy costs Cons: DC operator needs to build and maintain the plant Best places for a DC are often not the same as for plants Example: Apple is building a 20MW solar array in NC
Outline DC energy usage and carbon footprint Reducing carbon with renewables Our approach and research challenge Parasol: our solar micro-data center Software for leveraging solar energy Current and future works Conclusions
Our approach and challenge Self-generation with solar and/or wind Small and medium DCs Capital cost is lower Space required is smaller Easier to install and maintain
g CO2e per KWh over lifetime Solar and wind are clean 1000 900 800 700 600 500 400 300 200 100 0 [Sovacool 08] Massive environmental disruption Waste lasts for thousands of years
Solar is more available in the US Wind Solar
1997 1999 2001 2003 2005 2007 2009 2011 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 Cost of solar energy is decreasing 14 12 10 8 6 4 2 0 Grid electricity price (2011 c/kwh) 25 20 15 10 5 0 PV panel cost (2011 $/W) [EIA 12] [Solarbuzz 12]
1997 1999 2001 2003 2005 2007 2009 2011 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 Cost of solar energy is decreasing 14 12 10 8 6 4 2 0 Grid electricity price (2011 c/kwh) 25 20 15 10 5 0 PV panel cost (2011 $/W) [EIA 12] [Solarbuzz 12] Accounting for inverters, we may pay ~$5/W in NJ for Parasol With incentives, the cost may become ~$2/W
Main challenge: supply of power is variable! Mitigation approaches are not ideal Batteries and net metering We need to match the energy demand to the supply Many research questions: What kinds of DC workloads are amenable? What kinds of techniques can we apply? How well can we predict solar availability? Building hardware & software to answer questions
Outline DC energy usage and carbon footprint Reducing carbon with renewables Our approach Parasol: our solar micro-data center Software for leveraging solar energy Current and future works Conclusions
Research platform Parasol Solar-powered computing Software to exploit renewables within and across DCs Tradeoff between renewables, batteries, and grid energy Free cooling Wimpy servers Solid-state drives Full monitoring: resources, power, temperature, air
Parasol details Installed on the roof Steel structure Container to host the IT 16 solar panels: 3 kw Backup power Batteries: 32 kwh Power grid IT equipment 2 racks 64 Atom servers (so far): 1.7 kw 2 switches and 3 PDUs Cooling Free cooling: 110 W or 400 W Air conditioning: 2 kw Heating
You don t see this everyday
Container and solar panels
Electrical and cooling
IT equipment
Outline DC energy usage and carbon footprint Reducing carbon with renewables Our approach Parasol: our solar micro-data center Software for leveraging solar energy Current and future works Conclusions
Green DC software Smart energy and cost management GreenSlot [Supercomputing 11] Schedule batch jobs (SLURM) GreenHadoop [Eurosys 12] Schedule data-processing jobs (MapReduce)
Overall approach Predict green energy availability Weather forecasts Schedule jobs Maximize green energy use If green not available, consume cheap brown electricity May delay jobs but must meet deadlines Manage data availability if necessary Send idle servers to sleep to save energy
Nodes GreenSlot behavior Schedule: J1, J2 J2 J2 Power Now J1 J1 J2 Brown electricity price Job deadline Scheduling window Time
Nodes GreenSlot behavior Schedule: J3, J4 J2 J4 Power J1 J3 Now J3 J4 Brown electricity price Job deadline Time Scheduling window
Nodes GreenSlot behavior Schedule: J4 Weather prediction was wrong J2 J4 Power J1 J3 Now Brown electricity price Job deadline Scheduling window Time J4
Nodes GreenSlot behavior Schedule: J5 J2 J4 Power J1 J3 J5 Brown electricity price Job deadline Scheduling window Now Time J5
Median error (%) Energy (kwh) Energy prediction vs actual 2 1.5 1 0.5 0 Prediction Actual 20 15 10 5 0 0 6 12 18 24 30 36 42 48 Hours ahead
Median error (%) Energy (kwh) Energy prediction vs actual 2 1.5 1 0.5 0 cloud cover rain thunderstorm Prediction Actual 20 15 10 5 0 0 6 12 18 24 30 36 42 48 Hours ahead
GreenSlot Conventional GreenSlot for BSC workload 26 kwh 75 kwh $8.00 24% cost savings 38 kwh 63 kwh $6.06-24%
% GreenSlot for BSC workload 120 100 80 Most Best Average Worst 60 40 20 0 Green energy increase Cost savings
Outline DC energy usage and carbon footprint Reducing carbon with renewables Our approach Parasol: our solar micro-data center Software for leveraging solar energy Current and future works Conclusions
Current and future works DC placement with probabilistic guarantees GreenNebula Smart management of energy sources Green SLAs Tradeoff between performance and green energy use Collect and make sense of the monitoring data
Conclusions Topic is really exciting and has societal impact Lots left to do We would like to see more people working on it
Leveraging Renewable Energy in Data Centers Ricardo Bianchini Department of Computer Science Collaborators: Inigo Goiri, Jordi Guitart (UPC/BSC), Md. Haque, William Katsak, Kien Le, Thu D. Nguyen, and Jordi Torres (UPC/BSC)
Parasol: Lessons learned Not as easy as it seems Not as cheap as it seems Doing the complete homework (design) is not easy Hard to deal with facilities people (with exceptions!) Hard to deal with engineering companies (ditto) Placing it on the roof has extra cost, but is very cool Flexibility adds complexity Monitoring adds complexity Mistakes all around Delays, delays, and more delays