Top Banner
Clouds in High Energy Physics Randall Sobie Ins$tute of Par$cle Physics University of Victoria
31
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Clouds in High Energy

Clouds    in    

High  Energy  Physics  

Randall  Sobie  Ins$tute  of  Par$cle  Physics  

University  of  Victoria  

Page 2: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   2  

High  Energy  Physics  (Par$cle  Physics)  

 

The  area  of  physics  that  studies  the  fundamental  par$cles  of  nature  and  their  

interac$ons.  

SLAC  Linear  Accelerator  

Accelerators  

Sudbury  Solar  Neutrino  Detector  

Underground  labs  

Alpha  Magne$c  Spectrometer    Space  Sta$on    

Orbi;ng  labs  

Page 3: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   3  

Is  the  Higgs  boson  the  source  of  mass  of  our  fundamental  

par;cles?  

Page 4: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   4  

Why  is  the  universe  made    of  maCer    

and  not  equal  amounts  of  maCer/an;maCer?  

Page 5: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   5  

What  is  origin  of  Dark  MaCer  and  Dark  Energy?  

T  We  do  not  know  the  

composi$on  of    95%  of  the  universe  

Temperature  of  the  universe  WMAP  satellite  

Page 6: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   6  

Understanding  our  World  

Page 7: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   7  

Understanding  our  World  

Page 8: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   8  

Large  Hadron  Collider  at  CERN  in  Geneva    27  km  in  circumference                100  m  underground  

 Probes  distances  of  10-­‐20  m  

Page 9: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   9  

Large  Hadron  Collider  at  CERN  in  Geneva    27  km  in  circumference                100  m  underground  

 Probes  distances  of  10-­‐20  m  

Page 10: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   10  

ATLAS  Detector  

44  m  wide  25  m  diameter    

Page 11: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   11  

ATLAS  Detector  

44  m  wide  25  m  diameter    

Page 12: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   12  

ATLAS  Detector  

44  m  wide  25  m  diameter    

Page 13: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   13  

40  million  collisions  per  second  

100,000  collisions  selected  

200  events  per  second  

WLCG  Compu;ng  Grid  CERN  Tier  0  10  Tier1  sites  60+  Tier2  sites  

 140  PB  data  

Page 14: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   14  

Higgs  Discovery  in  2012    

The  ATLAS  and  CMS  experiments  see  evidence  for  a  Higgs-­‐like  par$cle    

Picture  shows  an  event  where  the  Higgs  candidate  decays  to  4  electron-­‐like  par$cles  

Page 15: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   15  

Number  of  candidates    (ver$cal  axis)  

 Mass  of  the  candidates  

(horizontal  axis)        

We  observe  an  excess  of  candidates  with  a  mass  of  

125  proton-­‐masses    

Search  for  Higgs  decays  to  4  “leptons”  (electrons  or  muons)    

Also  observed  in  the  CMS  experiment    

Page 16: Clouds in High Energy

Randall  Sobie        IPP/UVictoria   16  

July  4,  2012  

Page 17: Clouds in High Energy

Long-­‐term  preserva$on  of  so^ware  and  data  of  HEP  

experiments  

Clouds  in  High  Energy  Physics  

Randall  Sobie        IPP/UVictoria   17  

U$lize  special  compu$ng  resources  a_ached  to  the  

detectors  

Simplify  the  management  of  heterogeneous  in-­‐house  

resources  

Use  commercial  clouds  for  excep$onal  compu$ng  

demands  

Distributed  cloud  compu$ng  using  HEP  and  non-­‐HEP  

clouds  

Page 18: Clouds in High Energy

Using  Clouds  for  Data  Preserva$on  

Randall  Sobie        IPP/UVictoria   18  

SLAC  Linear  Accelerator   BaBar  Detector  

!"#$%&'()*+',--.

/+012'3"01#4'562%$7%8+%)"9%':4#;

!<==<'>'!?!?8

Electron-­‐positron  collision  

BaBar  experiment  stopped  recording  electron-­‐positron  collisions  in  2008  

Page 19: Clouds in High Energy

Using  Clouds  for  Data  Preserva$on  

Randall  Sobie        IPP/UVictoria   19  

SLAC  Linear  Accelerator   BaBar  Detector  

!"#$%&'()*+',--.

/+012'3"01#4'562%$7%8+%)"9%':4#;

!<==<'>'!?!?8

Electron-­‐positron  collision  

BaBar  experiment  stopped  recording  electron-­‐positron  collisions  in  2008  

Long  Term  Data  Access  system      

Sta$c  SL5  VMs    Full  access  to  BaBar  SW  and  data  

Transparent  to  the  user  In  opera$on  since  2010  

 h_p://today.slac.stanford.edu/feature/2010/babar-­‐prototype-­‐

data-­‐servers.asp  

 

Page 20: Clouds in High Energy

ATLAS/CMS  High  Level  Trigger  Clouds  

Randall  Sobie        IPP/UVictoria   20  

40  million  collisions  per  second  

100,000  collisions  selected  

200  events  per  second  

 HLT  Farms  

   

These  systems  are  used  in  real-­‐$me  when  there  is  colliding  beams  

 The  aim  is  to  use  the  resources  during  the  idle  periods  for  other  purposes  

 Enabled  as  private  OpenStack  clouds  

 See  talk  by  Toni  Perez    

Wednesday  1100  

 

50,000  cores    ATLAS,  CMS,  LHCb  

Page 21: Clouds in High Energy

Private  and  commercial  clouds  

Randall  Sobie        IPP/UVictoria   21  

Ibex  @  CERN  (J.  van  Eldik,  T.  Bell,  B.Moreira)    OpenStack  cloud  with  5000  cores    Provide  batch  services  and  cloud  services  

HEP  using  Amazon,  Google,  Rackspace  and  others    Star  Experiment  at  RHIC  (Brookhaven  NL)  Belle  Experiment  at  KEK  (Japan)  ATLAS  Experiment    Commercial  clouds  used  for  excep1onal  low  I/O  demands    Challenges:  iden1ty  management,  API  compa1bility,  VM  configura1on  and  network  connec1vity    Costs  are  higher  than  our  private  resources  

Page 22: Clouds in High Energy

Distributed  cloud  compu$ng  Grid  of  Clouds  

Randall  Sobie        IPP/UVictoria   22  

Seamlessly  use  mul$ple,  heterogeneous  IaaS  clouds  

for  batch  workloads  Sky  Compu;ng  K.  Keahey  etal  

Use  dedicated  HEP  and  non-­‐HEP  opportunis$c  resources    

Independent  of  the  IaaS  cloud  type      

Removes  any  applica$on  requirements  from  remote  site    

Support  mul$ple  projects  

Page 23: Clouds in High Energy

Distributed  cloud  overview  

Randall  Sobie        IPP/UVictoria   23  

Cloud  Scheduler  

 

IaaS  Clouds  

HTCondor    JobQueue  

UserJob  

User  VM  

Discover  User-­‐Job  

Page 24: Clouds in High Energy

Distributed  cloud  overview  

Randall  Sobie        IPP/UVictoria   24  

Cloud  Scheduler  

 

IaaS  Clouds  

HTCondor    JobQueue  

UserJob  

User  VM  

Boot  User-­‐VM  on  a  cloud  

Page 25: Clouds in High Energy

Distributed  cloud  overview  

Randall  Sobie        IPP/UVictoria   25  

Cloud  Scheduler  

 

IaaS  Clouds  

HTCondor    JobQueue  

UserJob  

User  VM  

VM  registers    with  HTCondor  

Page 26: Clouds in High Energy

Distributed  cloud  overview  

Randall  Sobie        IPP/UVictoria   26  

Cloud  Scheduler  

 

IaaS  Clouds  

HTCondor    JobQueue  

UserJob  

User  VM  

Dispatches  UserJob  to  VM  

Remote  VM  image  repository  (Nimbus  clouds)  Upload  VMs  to  OpenStack  clouds  Remote  data  repositories  for  all  clouds  

Page 27: Clouds in High Energy

Distributed  cloud  status  

Randall  Sobie        IPP/UVictoria   27  

Nimbus  Victoria(3)  O_awa  FG  Chicago  FG  SanDiego  FG  Florida  

OpenStack  Melbourne  CERN  Edmonton  Oxford  

Amazon  EC2  Google  GCE  

Opera;onal  since  Nov  2011    Approximately  250K  jobs  (Astronomy  500K  jobs)    Using  10  clouds  for  ATLAS  jobs  500-­‐1000  simultaneous  jobs  12  hour  jobs    ATLAS  jobs  submi_ed  from  CERN  

Page 28: Clouds in High Energy

Distributed  cloud  status  

Randall  Sobie        IPP/UVictoria   28  

Nimbus  Victoria(3)  O_awa  FG  Chicago  FG  SanDiego  FG  Florida  

OpenStack  Melbourne  CERN  Edmonton  Oxford  

Amazon  EC2  Google  GCE  

Opera;onal  since  Nov  2011    Approximately  250K  jobs  (Astronomy  500K  jobs)    Using  10  clouds  for  ATLAS  jobs  500-­‐1000  simultaneous  jobs  12  hour  jobs    ATLAS  jobs  submi_ed  from  CERN  

April  9  2013  8  clouds    100-­‐120  8-­‐core  VMs  

ç  

Melbourne  CERN-­‐-­‐Ibex  CANARIE  

WG-­‐Victoria  NRC-­‐O_awa  Victoria  

FG-­‐San  Diego  FG-­‐Chicago  Victoria  

April  8-­‐15  2013  100-­‐150  8-­‐core  VMs  

Page 29: Clouds in High Energy

Technology  innova$on  for  new  science  

Randall  Sobie        IPP/UVictoria   29  

The  OpenStack  developer  community  can  help  us    

Common  authen$ca$on      Centralized  VM  image  storage  

Consistent  meta-­‐data  Unique  cloud  names  

 Simplify  the  integraNon  of  OpenStack  clouds  for  us    

We  see  ourselves  as  integrators  rather  than  developers  of  cloud  technology  

Page 30: Clouds in High Energy

Summary  

Randall  Sobie        IPP/UVictoria   30  

High  Energy  Physics  

Understand  our  Universe  Studying  the  Higgs    

Search  for  Dark  Ma_er  Difference  between  an$ma_er/ma_er  

Impact  on  society  CERN  is  the  birthplace  of  the  WWW  Medical  physics  Technology  innova$on  HQP  

Compu;ng  Large  distributed  systems  Global  research  network  

Novel  use  of  compu$ng  technologies  

Page 31: Clouds in High Energy

Acknowledgements  

Randall  Sobie        IPP/UVictoria   31  

Contact  informa$on:        [email protected]  Web  sites:          h_p://rjs.phys.uvic.ca/  

             h_p://heprc.phys.uvic.ca/home/