Greening'Datacenters'Through'Self4 Genera5on'of'Renewable'Energy' - - PowerPoint PPT Presentation

▶
greening datacenters through self4 genera5on of renewable
SMART_READER_LITE
LIVE PREVIEW

Greening'Datacenters'Through'Self4 Genera5on'of'Renewable'Energy' - - PowerPoint PPT Presentation

Mar 05, 2024 331 likes •807 views

Greening'Datacenters'Through'Self4 Genera5on'of'Renewable'Energy' Thu'D.'Nguyen' Department'of'Computer'Science' ' Collaborators:'Ricardo'Bianchini,' Inigo'Goiri,'Md.'Haque,'William'Katsak,'Kien'Le' ' Mo5va5on'

slide-1
SLIDE 1

Greening'Datacenters'Through'Self4 Genera5on'of'Renewable'Energy'

Thu'D.'Nguyen' Department'of'Computer'Science' ' Collaborators:'Ricardo'Bianchini,' Inigo'Goiri,'Md.'Haque,'William'Katsak,'Kien'Le' '

slide-2
SLIDE 2

Mo5va5on'

  • Datacenters'consume'massive'amounts'of'energy'(electricity)'
  • Vast'majority'currently'due'to'small'and'medium'scale'

datacenters'

0' 30' 60' 90'

2000' 2005' 2010' Billion'KWh/year' Electricity'consump5on'of'US'DCs'[JK’11]'

0' 30' 60' 90' 120' 150' 180' 210' 240' 270'

2000' 2005' 2010' Billion'KWh/year' Electricity'consump5on'of''WW'DCs'[JK’11]' 2%' 1.5%'

Rutgers'University' 2'

slide-3
SLIDE 3

Mo5va5on'

  • Electricity'comes'mostly'from'burning'fossil'fuels'

CO2'of'world4wide'DCs'[Mankoff’08]' Electricity'sources'in'US'&'WW'[DOE’10]'

100' 104' 108' 112' 116' 120'

Nigeria' Data'Centers' Czech'Rep.'

35th' 34th' MMT/year'

0%' 20%' 40%' 60%' 80%' 100%'

US' World'

Others' Renewables' Nuclear' Natural'Gas' Coal'

Rutgers'University' 3'

Can'we'use'renewables'to'reduce'this'footprint?'

  • Reducing'electricity'costs'would'be'nice'too'
slide-4
SLIDE 4

Outline'

  • Mo5va5on:'DC'energy'usage'and'carbon'footprint'
  • Reducing'carbon'footprint'&'cost'with'renewables'
  • Our'target'and'research'challenges'
  • Parasol:'our'solar4powered'micro4datacenter'
  • GreenSwitch:'managing'power'sources'
  • Previous,'current,'and'related'work'
  • Conclusions'

Rutgers'University' 4'

slide-5
SLIDE 5

Greening'Datacenters'

  • Power'purchase'agreement,'off4site'genera5on'

– Renewable'energy'produced'at'the'best'loca5on' – Requires'transmission'

  • Construc5on'of'transmission'lines'
  • Energy'losses:'~15%'[IEC’07]'
  • Variability'challenging'for'grid'operators'

– Example:'Google'buys'wind'power'from'NextEra'

  • Self4genera5on,'co4loca5on'

– Lower'peak'power,'energy'costs'with'self4genera5on' – Loca5on'may'not'be'ideal'for'DC'or'renewable'plant' – Examples:'

  • Microsoj'placed'DC'near'a'hydro'plant'in'OR'
  • Apple'built'a'40MW'solar'array'in'NC'
  • No'approach'is'perfect'

Rutgers'University' 5'

slide-6
SLIDE 6

Self4Genera5on'Example:'Apple'NC'

Apple'DC'in'Maiden,'NC' 40MW'solar'farm'

6' Rutgers'University'

slide-7
SLIDE 7

Outline'

  • DC'energy'usage'and'carbon'footprint'
  • Reducing'carbon'footprint'&'cost'with'renewables'
  • Our'target'and'research'challenges'
  • Parasol:'our'solar4powered'micro4datacenter'
  • GreenSwitch:'managing'power'sources'
  • Previous,'current,'and'related'work'
  • Conclusions'

Rutgers'University' 7'

slide-8
SLIDE 8

Our'Research'Target'

  • Self4genera5on'or'co4loca5on'with'solar'and/or'wind'

– Pros:'Clean'and'available' – Cons:'Space'and'cost'

Rutgers'University' 8'

slide-9
SLIDE 9

Solar'and'Wind'Are'Clean'

0' 100' 200' 300' 400' 500' 600' 700' 800' 900' 1000' g'CO2e'per'KWh'over'life5me' [Sovacool’08]'

Rutgers'University' 9'

slide-10
SLIDE 10

0' 100' 200' 300' 400' 500' 600' 700' 800' 900' 1000'

Solar'and'Wind'Are'Clean'

g'CO2e'per'KWh'over'life5me' [Sovacool’08]'

Rutgers'University' 10'

slide-11
SLIDE 11

Solar'More'Available'In'US'

Wind' Solar' [NREL’12]'

Fair' Good' Excellent' Outstanding' Superb'

Rutgers'University' 11'

slide-12
SLIDE 12

Solar'PV'Efficiencies'Are'Increasing'

[IEA’10]' Efficiency'rates'of'PV'modules'

Rutgers'University' 12'

slide-13
SLIDE 13

0' 5' 10' 15' 20' 25'

Solar'PV'Capacity'Factors'Today'

[PVOutput’12]'

Rutgers'University' 13'

slide-14
SLIDE 14

Cost'of'Solar'PV'Energy'Decreasing'

Grid'electricity'prices'have'been'increasing:'30%+'since'1998'[EIA’12]'

[DOE’11,Solarbuzz’12]' 0' 4' 8' 12' 16' 20' 1985' 1987' 1989' 1991' 1993' 1995' 1997' 1999' 2001' 2003' 2005' 2007' 2009' 2011' 2013' 2015' 2017' 2019' 2021' 2023' 2025' 2027' 2029' 2011'Dollars'per'Wao'

Inverters' Panels' Installed'

Rutgers'University' 14'

slide-15
SLIDE 15

0' 4' 8' 12' 16' 20' 1985' 1987' 1989' 1991' 1993' 1995' 1997' 1999' 2001' 2003' 2005' 2007' 2009' 2011' 2013' 2015' 2017' 2019' 2021' 2023' 2025' 2027' 2029' [DOE’11,Solarbuzz’12]' 2011'Dollars'per'Wao'

Inverters' Panels' Installed'

Cost'of'Solar'PV'Energy'Decreasing'

spike'in'demand' world4wide'recession' back'to'historical'levels'

Rutgers'University' 15'

slide-16
SLIDE 16

0' 4' 8' 12' 16' 20' 1985' 1987' 1989' 1991' 1993' 1995' 1997' 1999' 2001' 2003' 2005' 2007' 2009' 2011' 2013' 2015' 2017' 2019' 2021' 2023' 2025' 2027' 2029' [DOE’11,Solarbuzz’12]' 2011'Dollars'per'Wao'

Inverters' Panels' Installed'

Cost'of'Solar'PV'Energy'Decreasing'

<'1/2'of'current'cost'

With'incen5ves,'the'installed'price'can'go'down'by'another'40460%'

Rutgers'University' 16'

slide-17
SLIDE 17

Solar'Space'and'Cost:'Present'and'Future'

Space&as&a&factor&of&rack&area& Present& Future&(202042030)& Density'per'rack' 8kW'(200W'1U'servers)' ~47x' ~24x' 2kW'(25W'0.5U'servers)' ~12x' ~6x'

Assuming'30%'server'u5liza5on,'50%'solar'energy,'NJ'capacity'factor,'and'1'row'of'panels'

Rutgers'University' 17'

slide-18
SLIDE 18

Solar'space'and'cost:'Present'and'future'

Space&as&a&factor&of&rack&area& Present& Future&(202042030)& Density'per'rack' 8kW'(200W'1U'servers)' ~47x' ~24x' 2kW'(25W'0.5U'servers)' ~12x' ~6x' Cost&per&AC&Wa:& Present& Future&(202042030)& ~$2.30' <'$1.20' Time&to&amor>ze&cost& Present& Future&(202042030)& ~12'years' <'6'years'

Assuming'30%'server'u5liza5on,'50%'solar'energy,'NJ'capacity'factor,'and'1'row'of'panels' Assuming'above'costs,'NJ'capacity'factor,'and'NJ'grid'energy'prices' Assuming'self4genera5on'and'federal'+'state'incen5ves'

Rutgers'University' 18'

slide-19
SLIDE 19

Solar'space'and'cost:'Present'and'future'

Space&as&a&factor&of&rack&area& Present& Future&(202042030)& Density'per'rack' 8kW'(200W'1U'servers)' ~47x' ~24x' 2kW'(25W'0.5U'servers)' ~12x' ~6x'

Assuming'30%'server'u5liza5on,'50%'solar'energy,'NJ'capacity'factor,'and'1'row'of'panels'

Wind'takes'~12x'less'space'and'is'~3x'cheaper'

Cost&per&AC&Wa:& Present& Future&(202042030)& ~$2.30' <'$1.20' Time&to&amor>ze&cost& Present& Future&(202042030)& ~12'years' <'6'years'

Assuming'above'costs,'NJ'capacity'factor,'and'NJ'grid'energy'prices' Assuming'self4genera5on'and'federal'+'state'incen5ves'

Rutgers'University' 19'

slide-20
SLIDE 20

Main'Challenge:'Power'Supply'is'Variable''

  • Power'genera5on'is'variable'

– Unlikely'to'match'workload'

  • Match'power'demand'and'supply'

Power' Time' Load' Variable'

20'

Solar'power' Workload' Source?' Storage?'

Rutgers'University'

slide-21
SLIDE 21

Main'Challenge:'Power'Supply'is'Variable''

  • Many'research'ques5ons:'

– What'kinds'of'DC'workloads'are'amenable?' – What'kinds'of'techniques'can'we'apply?' – How'well'can'we'predict'solar'energy'availability?' – If'baoeries'are'available,'how'should'we'manage'them?' – Can'we'leverage'geographical'distribu5on?'

  • Building'hardware'&'sojware'to'answer'ques5ons'

Rutgers'University' 21'

slide-22
SLIDE 22

Outline'

  • DC'energy'usage'and'carbon'footprint'
  • Reducing'carbon'footprint'&'cost'with'renewables'
  • Our'target'and'research'challenges'
  • Parasol:'our'solar4powered'micro4datacenter'
  • GreenSwitch:'managing'power'sources'
  • Previous,'current,'and'related'work'
  • Conclusions'

Rutgers'University' 22'

slide-23
SLIDE 23

The'Rutgers'Parasol'Project'

Rutgers'University' 23'

slide-24
SLIDE 24

Parasol:'Our'Hardware'Prototype'

  • Unique'research'plarorm'

– Solar4powered'compu5ng' – Remote'DC'deployments' – Sojware'to'exploit'renewables'within'and'across'DCs' – Tradeoff'between'renewables,'baoeries,'and'grid'energy' – Free'cooling,'wimpy'servers,'solid4state'drives'

Rutgers'University' 24'

slide-25
SLIDE 25

Parasol:'Our'Hardware'Prototype'

Rutgers'University' 25'

slide-26
SLIDE 26

Parasol'Details'

  • Steel'structure'on'the'roof'

– 16'solar'panels:'3.2'kW'peak' – Container'hosts'2'racks'of'IT'

  • Backup'power'

– Baoeries:'32kWh' – Grid'

  • IT'equipment'

– 64'Atom'servers'(so'far):'1.9'kW' – 2'switches'

  • Cooling'

– Free'cooling:'10'−'400'W' – Air'condi5oning:'2'kW'

Rutgers'University' 26'

slide-27
SLIDE 27

Electrical'infrastructure'

Charge' Controller' Inverter' Electrical' Panel' DC' AC' AC' DC' AC' Power'grid' AC'

27'

IT'Equipment'

Rutgers'University'

slide-28
SLIDE 28

Outline'

  • DC'energy'usage'and'carbon'footprint'
  • Reducing'carbon'footprint'&'cost'with'renewables'
  • Our'target'and'research'challenges'
  • Parasol:'our'solar4powered'micro4datacenter'
  • GreenSwitch:'managing'power'sources'
  • Previous,'current,'and'related'work'
  • Conclusions'

Rutgers'University' 28'

slide-29
SLIDE 29

Example'Energy'Source'Management'

Charge' Controller' Inverter' Electrical' Panel' DC' AC' AC' DC' AC' Power'grid' AC' Time' Power'

29'

IT'Equipment'

Rutgers'University'

slide-30
SLIDE 30

Example'Energy'Source'Management'

Charge' Controller' Inverter' Electrical' Panel' DC' AC' AC' DC' AC' Power'grid' AC' Time' Power'

30'

IT'Equipment'

Rutgers'University'

slide-31
SLIDE 31

Example'Energy'Source'Management'

Charge' Controller' Inverter' Electrical' Panel' DC' AC' AC' DC' AC' Power'grid' AC' Time' Power'

31'

IT'Equipment'

Rutgers'University'

slide-32
SLIDE 32

Possible'Energy'Source'Management'

Power' Time'

Basic'

32' Rutgers'University'

slide-33
SLIDE 33

GreenSwitch'

  • Minimize'brown'electricity'cost'

– Brown'energy' – Peak'brown'power' – Baoery'life5me'constraint'

  • Manage'energy'sources'

– Use'solar/net'metering' – Charge/discharge'baoery' – Limit'brown'peak'power'

  • Manage'workload'

– Turn'servers'on/off' – Delay'deferrable'jobs'

Parasol' GreenSwitch'

Perform'changes' Get'status'

33' Rutgers'University'

slide-34
SLIDE 34

GreenSwitch'

GreenSwitch'Architecture'

Predictor' Baoery' Charge'Level' Workload' Predic5on' Energy' Availability' Predic5on' Solver' Energy'Source' Schedule' Workload' Schedule' Configurer'

Parasol'

Time' Power' Brown' Price' Solar' Time' Power' Workload' Use'solar'to'power'all'servers' Charge'baoery'with'surplus'solar' Workload' Model'&' MILP'

34' Rutgers'University'

slide-35
SLIDE 35

Experimental'Environment'

  • Evalua5on'on'64'Parasol'nodes'

– 12'one4day'experiments' – Deferrable'vs.'non4deferrable'workloads' – Baseline'datacenter'(no'solar,'no'baoeries,'no'delays)'

  • New'Jersey'brown'electricity'pricing'

– On/off4peak'energy,'peak'power,'net'metering'

  • GreenSwitch'for'Hadoop'(configurer)'

35' Rutgers'University'

slide-36
SLIDE 36

Experimental'Environment'

36'

SWIM:'Facebook'based'workload'[MASCOTS’11]'

IT'load'

Rutgers'University'

slide-37
SLIDE 37

Parasol'Without'GreenSwitch'

Green'use' Green'available' Net'metering' Brown'use' IT'load'

66%'cost'savings'→'Solar'amor5zed'in'7'years'

37' Rutgers'University'

slide-38
SLIDE 38

GreenSwitch:'Non4Deferrable'Workload'

Baoery'discharge' Baoery'charge' IT'load'

75%'cost'savings'→'Baoeries'cannot'be'amor5zed'

Peak'grid'power'

38'

Green'use' Brown'use' Green'available'

Rutgers'University'

slide-39
SLIDE 39

GreenSwitch:'Deferrable'Workload'

Baoery'discharge' Baoery'charge' IT'load' Net'metering'

96%'cost'savings'→'Solar'+'baoeries'amor5zed'in'7.6'years''

39'

Green'available' Green'use'

Rutgers'University'

slide-40
SLIDE 40

Parasol:'A'Real'System'

  • Real'sojware'running'on'real'hardware'
  • Power'losses'
  • Overhead'of'energy'source'switching'
  • System'limita5ons'

– Net'metering'vs.'baoery'charging' – Use'grid'vs.'net'metering' – Green'baoery'charging'vs.'use'grid'

Rutgers'University' 40'

slide-41
SLIDE 41

Outline'

  • DC'energy'usage'and'carbon'footprint'
  • Reducing'carbon'footprint'&'cost'with'renewables'
  • Our'target'and'research'challenges'
  • Parasol:'our'solar4powered'micro4datacenter'
  • GreenSwitch:'managing'power'sources'
  • Previous,'current,'and'related'work'
  • Conclusions'

Rutgers'University' 41'

slide-42
SLIDE 42

Other'Works'on'Green'DC'Sojware'

  • Follow'the'renewables'[HotPower’09,'IGCC’10]'
  • Delay'batch'jobs'while'respec5ng'deadlines'[SC’11,'

Eurosys’12]'

  • Power'source'management'[SustainIT’12,'

ASPLOS’13]'

  • Green'SLAs'in'HPC'clouds'[IGCC’13]'

Rutgers'University' 42'

slide-43
SLIDE 43

Current'Works'

  • Temperature4'and'varia5on4aware'management'of'

free4cooled'datacenters'

  • DC'placement'world4wide'for'cost4effec5ve'follow4

the4renewables'

  • Matching'power'demand'to'power'supply'for'non4

deferrable'workloads'

– Trading4off'response'5me'and'durability'for' interac5ve'workloads'(GreenCassandra)'

Rutgers'University' 43'

slide-44
SLIDE 44

Related'Work'

  • Blink,'UMass,'Amherst'
  • Algorithms'for'Sustainable'IT,'Caltech'
  • Sustainable'datacenters,'HP'Labs''
  • Ren'et'al.,'MASCOTS'2012'
  • Li'et'al.,'ISCA'2012'

Rutgers'University' 44'

slide-45
SLIDE 45

Conclusions'

  • Greening'datacenters'

– Challenges'&'opportuni5es' – Hardware/sojware'solu5on'

  • GreenSwitch'benefits'

– Delaying'load'and'solar'gives'the'best'results' – Reduces'amor5za5on'5me'by'1.842x' – Flexibility:'no'baoeries,'workloads,'wind'…'

45' Rutgers'University'

slide-46
SLIDE 46

Rutgers'University' 46'