Proactive and reactive thermal optimization techniques to improve energy efficiency in data centers

96e2d41af0b2b857d8ec7cd65d170e72?s=47 GreenLSI
February 14, 2013

Proactive and reactive thermal optimization techniques to improve energy efficiency in data centers

Marina Zapater presents her work at the PICATA Workshop. This workshop is intended to know the diverse groups of people recently incorporated thank to PICATA programme of Moncloa campus and who are researching and assessing the clusters.

The Program for International Talent Recruitment (PICATA) has focused on bringing in students and researchers from all over the world, in a determined effort towards internationalization and talent recruitment with different actions. The PICATA Programme offers sholarships for the development of PhD thesis marked by at least two practising doctors from the two associated Universities, the UCM and the UPM, with the possibility of participation by doctors from the other associated Institutions within the context of the Campus Moncloa in these areas: Global Change and New Energies, Materials for the Future, Agri-food and Health, Innovative Medicine, and Heritage.

96e2d41af0b2b857d8ec7cd65d170e72?s=128

GreenLSI

February 14, 2013
Tweet

Transcript

  1. Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   1  

    MONCLOA  Campus  of  Interna4onal   Excellence   Proactive and Reactive Thermal Optimization Techniques to Improve Energy Efficiency in Data Centers Workshop  PICATA   Marina  Zapater   José  L.  Ayala,  José  M.  Moya  
  2. Laboratorio  de  Sistemas  Integrados  (LSI)   Departamento  de  Ingeniería  Electrónica

      ETSI  Telecomunicación   Universidad  Politécnica  de  Madrid   ArTeCS  Group   Group  of  Architecture  and  Technology  of  CompuQng  Systems   Facultad  de  InformáQca   Universidad  Complutense  de  Madrid   Presentation Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   2  
  3. Data Centers Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013  

    3  
  4. Motivation Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   4

      •  Power  consumpQon  in  data  centers   –  1.3%  world  energy  producQon  in  2010   –  USA:  80  billion  KWh/year  in  2011  =  1.5xNYC   –  250  billion  KWh/year  in  2010   •  More  than  43  Million  tons  of  CO2  /  year   •  More  water  than  paper,  automoQve,  petrol,  wood   or  plasQc  industry                      Jonathan  Koomey.  2011.  Growth  in  Data  center  electricity  use  2005  to  2010  
  5. Motivation Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   5

      •  It  is  expected  for  total  data   center  electricity  use  to   exceed  400  GWh/year  by   2015.   •  The  required  energy  for   cooling  will  conQnue  to  be  at   least  as  important  as  the   energy  required  for  the   computaQon.   •  Energy  op4miza4on  of   future  data  centers  will   require  a  global  and  mulQ-­‐ disciplinary  approach.   0   5000   10000   15000   20000   25000   30000   35000   2000   2005   2010   High-­‐end  servers   Mid-­‐range  servers   Volume  servers   0   50   100   150   200   250   300   2000   2005   2010   Infrastructure   CommunicaQons   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)   World  server  installed  base  (thousands)   Electricity  Use  (billion  KWh/year)  
  6. State of the Art Energy Savings for different abstraction levels

    AbstracQon  level   •  Higher  levels  of   abstracQon  bring   more  benefits   •  ApplicaQon-­‐level   sQll  has  to  be   explored.   SoluQons  proposed  by  the  State  of  the  Art   6   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013  
  7. Our perspective   •  Using  the  knowledge  about  the  energy

     demand  of  the   applica4ons,  the  features  of  the  computa4on  and   cooling  resources  to  apply  proacQve  opQmizaQon   techniques   •  Global  strategy  to  integrate  mulQple  informaQon   sources  and  coordinate  decissions  to  reduce  overall   power  consumpQon.   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   7   Proactive and reactive holistic approach
  8. Our perspective •  State  of  the  Art:    PUE  ≈

     1,2   – The  important  part  is  IT  energy  consumpQon   – Current  work  is  focused  on  decreasing  PUE   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   8   IT and Cooling power cooling IT TOTAL P P P + = IT TOTAL P P PUE =
  9. Our perspective Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013  

    9   IT and Cooling power cooling IT TOTAL P P P + = IT TOTAL P P PUE = •  Minimize  IT  power   •  Jointly  minimize  IT  and  cooling  
  10. Minimizing IT Power Leveraging heterogeneity •  Usage  heterogeneity  (existance  of

     different  servers)  to   minimize  energy  consumpQon:   –  StaQc:  Finding  the  best  data  center  set-­‐up,  given  a  number  of   heterogeneous  machines   –  Dynamic:  opQmizaQon  of  task  allocaQon   M.  Zapater,  J.M.  Moya,   J.L.  Ayala.  Leveraging   Heterogeneity  for   Energy  MinimizaQon  in   Data  Centers,  CCGrid   2012   10   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013  
  11. Minimizing IT power Application Awareness WORKLOAD   Scheduler   Resource

        Manager   Execu4on   11   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013  
  12. Heterogeneity Application Awareness WORKLOAD   Scheduler   Resource    

    Manager   Execu4on   Profiling  and   Classifica4on   Energy     Op4miza4on   12   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013  
  13. Cooling management •  Control  of  the  fan  speed  of  a

      server     –  Enterprise  server:  Sparc  T3  (256   threads)   –  Real  measures  with  server   internal  sensors   •  We  can  find  an  opQmum   pointbetween  leakage  and   cooling  to  minimize  power   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   13   Work  in  collaboraQon  with:     Leakage-cooling tradeoffs at the server level
  14. Cooling management Fig. 4. Test 3 temperature sensor readings for

    the three different controlle 0.1 0.2 ) Energy difference between 1800RPM and 2400RPM for clustered allocation analytical model for leakag fan speeds for varying utiliza 4. Test 3 temperature sensor readings for the three different controllers 0RPM for clustered allocation analytical model for leakage power and find the optimum fan speeds for varying utilization values. Based our analytical Work  in  collaboraQon  with:     Leakage-cooling tradeoffs at the server level 14   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013  
  15. Cooling & IT Joint Opt. •  Deriving  a  data  room

     thermal   model  to  jointly  allocate   computaQonal  and  cooling   resources   –  Gathering  environmental  data   through  sensors  (WSN)   –  Server  sensors   –  Workload  informaQon   •  Usage  of  geneQc  programming   and  geneQc  algoQthms   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   15   Work in Progress
  16. Holistic aproach •  System  that  increases  the  knowledge  of  the

     data  center   •  Real  implementaQon  scenario  at  CeSViMa   •  Power  opQmizaQon  in  globally  distributed  systems     Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   16   Proactive and reactive techniques Data  Center   state   Op4miza4on   Datacenter   Sensing   Decission   proposal   GreenDISC  Project:  HW/SW   Technologies  for  Energy   Efficiency  in  Distributed   Compu4ng  Systems.   UCM-­‐UPM     TEC2012-­‐33892.   Spanish  Ministry  of  Economy  and   CompeQQveness  
  17. Research goals •  Energy  and  CO2  carbon  footprint  reducQon  in

      data  centers   – So  far,  25%  reducQon  in  IT  resource  management   opQmizaQons   – 10%  energy  reducQon  in  fan  control  policies   •  Joint  IT/cooling  techniques  are  expected  to   bring  much  more  benefits.   •  SoluQons  in  a  real  environment:  CeSViMa     Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   17   Expected impact
  18. Questions? Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   18

      Thank  you  for    your  a^en4on   Marina  Zapater   marina@die.upm.es   B105.  ETSI  Telecomunicación   Avda  Complutense,  30   91  549  57  00  (+  4227)   hsp://greenlsi.die.upm.es   hsp://artecs.dacya.ucm.es/    
  19. Research results •  M.  Zapater,  J.L.  Ayala,  J.M.  Moya,  K.

     Vaidyanathan,  K.  Gross,  A.K.  Coskun,  Leakage  and  Temperature   Aware  Server  Control  for  Improving  Energy  Efficiency  in  Data  Centers.  To  apper  in:  DATE’13,  2013.     •  M.  Zapater,  J.  L.  Ayala,  and  J.  M.  Moya.  “GreenDisc:  a  HW/SW  energy  opQmizaQon  framework  in  globally   distributed  computaQon”,  J.  Bravo,  D.  López-­‐de  Ipiña,  and  F.  Moya,  Ed.,  Springer  Berlin  Heidelberg,  2012,   pp.  1-­‐8.     •  M.  Zapater,  J.  L.  Ayala,  and  J.  M.  Moya,  “Leveraging  heterogeneity  for  energy  minimizaQon  in  data   centers”,  in  Proceedings  of  the  2012  12th  IEEE  InternaQonal  Symposium  on  Cluster,  Cloud  and  Grid   CompuQng  (CCGRID  2012),  Washington,  DC,  USA,  2012.   •  M.  Zapater,  C.  Sanchez,  J.  L.  Ayala,  J.  M.  Moya,  and  J.  L.  Risco-­‐Maryn,  “Ubiquitous  green  compuQng   techniques  for  high  demand  applicaQons  in  smart  environments”  Sensors,  vol.  12,  iss.  8,  pp.  10659-­‐10677,   2012   •  M.  Zapater,  P.  Arroba,  J.  M.  Moya,  and  Z.  Bankovic,  “A  State-­‐of-­‐the-­‐Art  on  energy  efficiency  in  today’s   datacentres:  researcher’s  contribuQons  and  pracQcal  approaches”,  UPGRADE,  vol.  12,  iss.  4,  pp.  67-­‐74,   2011.   •  M.  Zapater,  J.  L.  Risco,  J.  L.  Ayala,  and  J.  M.  Moya,  “Combined  Dynamic-­‐StaQc  approach  for  Thermal-­‐ Awareness  in  heterogeneous  data  centers”  IWIA  2010.   Marina  Zapater  |  Workshop  PICATA  |  14-­‐02-­‐2013   19