Risk based QA af Michael Agerkvist Pedersen, Radiometer Medical

100 %
0 %
Information about Risk based QA af Michael Agerkvist Pedersen, Radiometer Medical

Published on March 12, 2014

Author: InfinITnetvaerk

Source: slideshare.net


Oplægget blev holdt ved et seminar i InfinIT-interessegruppen Processer & IT Nord den 5. marts 2014. Læs mere om interessegruppen her: http://infinit.dk/dk/interessegrupper/processer_og_it/processer_og_it.htm

07/03/14   1   Risk  Based  QA   Michael  Agerkvist  Petersen   miap@post3.tele.dk   dk.linkedin.com/in/michaelagerkvist   Michael  Agerkvist  Petersen   •  QA  @  Radiometer  Medical   •  18+  years  with  Medical  Devices   – HW  development   – SW  development   – Project  Management   – Process  Improvement   – QA   •  Owner  of  MDCA  (Medical  Device  Compliance   Assistance)  –  Spare  Qme  job  J  

07/03/14   2   Risk  Based  QA   •  Based  on  my  experience  from:   –  Making  a  class  C  Medical  Device  at  Novo  Nordisk   –  Working  with  Encrypted  Pin  Pads  (ATM  keyboards)  at   Cryptera   –  Other  projects     •  This  will  not  be  a  complete  introducQon  to   (safety)  risk  management  or  Risk  Based  TesQng.   Agenda   •  IntroducQon   •  Risk  Based  Quality  Assurance   – Regulatory  Risks  –  Process  Rigor   – Risk  Based  DocumentaQon  Rigor   – Risk  Based  TesQng   – ProacQve  QA   •  Examples   •  PiZalls  

07/03/14   3   IntroducQon   Some  definiQons   •  Quality:  The  totality  of  features  and  characterisQcs  of  a   product  or  service  that  bears  its  ability  to  saQsfy  stated   or  implied  needs  [ISO]   –  There  are  many  different  customers:  Users,  OrganizaQon,   Regulatory….   •  QA:  Quality  Assurance  –  many  ways  to  implement  this   in  pracQce     –  Ensuring  that  development  is  in  compliance  (with  defined   process)   –  Doing  the  actual  tesQng   –  “anything”  which  helps  ensuring  Quality  to  the  different   customers  

07/03/14   4   Soeware  today   •  Today  soeware  controls  more  and  more  in  our   daily  life.   •  Soeware  failures  may  negaQvely  affect   business,  human  health  or  even  human  life   •  ResulQng  in  increasing  public  and  regulatory   demands  for  befer  products  and  even  more   Qme  pressure.   Regulatory  Compliance   •  More  and  more  industries   gets  regulated   •  Some  are  more  regulated   than  others   Accident   Injury  or   other   loss   Public   reacQon   New  laws   &   regulaQons   Regulated  Industry   Nuclear   Flight   Medical  Device   Military   Payment  systems   Public  Systems   •  The  bar  is  raised  over  Qme   [Not  to  scale]  

07/03/14   5   TesQng  Paradox   •  TesQng  is  a  structured  approach  to  reduce  the  number  of  defects  it   is  the  last  acQvity  before  release.  So  it  is  oeen  the  first  to  be   sacrificed     •  TesQng  is  always  a  sample.  You  can  never  test  everything,  and  you   can  always  find  more  to  test.     •  A  good  test  case  is  one  finding  a  defect  –  running  a  complete  test   suite  without  finding  a  single  defects  will  not  add  much  value.   •  The  problem  with  most  systemaQc  test  methods,  like  black  box   methods    (equivalence  parQQoning,  boundary  value  analysis  ,   cause-­‐effect  graphing,  etc.),  is  that  they  generate  too  many  test   cases,  many  of  them  will  never  find  a  defect.     •  However  –  in  a  regulated  world  it  adds  value  to  run  tests  which   does  not  find  defects  –  a  “clean  sheet”  is  needed  to  make  a   submission  without  too  many  quesQons   Risk  Based  Approach   •  A  way  to  lessen  the  the  work  load.  Could  be  to   test  more  in:   – bad  areas  of  the  product.     – the  most  important  funcQonal  areas  and  product   properQes.     •  And  tesQng  less  in  the  other  areas…   •  But,  how  to:   – find  the  right  areas?   – PrioriQzing  other  QA  acQviQes?    

07/03/14   6   Risk  Based  Quality  Assurance   TradiQonal  QA   •  Consist  of:   –  Review   –  Dynamic  Test   –  StaQc  analysis   –  Templates   –  Source  Code  standards   –  Traceability  analysis   –  Checklists   –  Etc   •  They  are  typically:   –  Time  consuming   –  Passive   –  Always  compromised  due  to  Schedule  pressure  

07/03/14   7   Balancing  QA   Too  li-le   •  Safety  risk   •  Poor  reliability   •  AddiQonal  cost   •  Project  delay   •  Regulatory  failure   •  Customer  complaints       Too  much   •  AddiQonal  cost   •  Project  delay   •  Under  verificaQon  (in  the  more   important  areas)   •  Increased  maintenance         Result  of  poor  QA   Liability  and  liQgaQon     Recalls     Loss  of  company  image     In-­‐market  updates     ReducQon  in  performance     No  regulatory  approval     Project  delay   Death     Major  Injury     Minor  Injury     Security  violaQon     ReducQon  in  performance     Loss  of  data   Business  cost   Customer  cost   Severity  High  Low  

07/03/14   8   Risk  Based  QA  Approach   •  Some  QA  acQviQes  will  sQll  be  Passive   –  They  are  well  known  and  needed   •  Some  QA  acQviQes  will  be  more  ProacQve   –  We  know  we  are  going  to  release  with  defects  -­‐  so  why  not  try  to   miQgate  the  impact  of  specific  types  of  defects  in  the  design?   •  All  QA  acQviQes  will  be  prioriQzed   –  Some  defects  or  lack  of  process/documentaQon  will  do  more  harm   than  others.  So  why  use  the  same  effort  and  acQviQes  on  every  feature   or  enQty?     •  Use  the  risk  based  QA  approach  to     –  Find  the  most  criQcal  defects  as  early  as  possible  at  the  lowest  effort/ cost   –  Find  the  most  criQcal  classes  of  defects  and  miQgate  the  impact  of   them  in  the  design   –  Balance  the  rigor  of  process,  documentaQon  and  design   Some  risk  definiQons   •  Harm:  Physical  injury  or  damage  to  the  health  of  people,  or   damage  to  property  or  the  environment.     –  From  project  delay  (financial  loss)  to  death  (e.g.  of  user)   •  Hazard:  PotenQal  source  of  harm   •  Probability:  Of  a  harm  to  occur.     –  Could  correspond  to  the  frequency  of  funcQonality  usage  by  the   user.   •  Severity:  Measure  of  the  possible  consequence  of  a  hazard   –  Customer  cost:  Loss  of  data  -­‐>  Security  violaQon  -­‐>  Injury-­‐ >Death   –  Business:  Project  delay-­‐>Recalls-­‐>Liability&liQgaQon   •  Risk  =  Severity  x  Probability   [IEC  14971]  

07/03/14   9   Probability   •  Probability  of  failure   – Usage  frequency  (funcQons  used  several  Qmes  a   day  vs  once  in  lifeQme)   – For  Medical  Device  Soeware,  probability  for  SW   defects  =  100%  -­‐  but  the  probability  of  Harm   needs  to  take  probabiliQes  from  the  enQre  chain   of  events   Probability  Levels   Probability  of  Harm Probability  of  Harm  Descrip@on   Ra@ng DefiniQon 5 Frequent   Constantly  present 4 Probable Have  been/  will  be  reported  but  no  more  than  once  per   month 3 Occasional Have  been/  will  be  reported  but  no  more  than  once  per   year 2 Remote Have  been/  will  be  reported  but  no  more  than  once  in   products  lifeQme  (~10  years) 1 Improbable Considered  unlikely  to  occur This  is  not  science…  And  very  difficult  in  real  life…  So  do  spend   too  much  Qme  finding    the  right  value  

07/03/14   10   Severity  Levels   Severity  of  poten@al  Harm Harm  descrip@on   # Term Descrip@on 5 Catastrophic Results    in  immediate  death  of  paQent  or  user   •  Immediate  death  of  person  caused  by  Explosion,  Fire  or   Electrical  shock   4 Serious Results  in  permanent  impairment   or     criQcal  injury  that  would  require  medical  or   surgical  intervenQon  to  preclude  irreversible   impairment  or  damage   •  Incorrect  medical  treatment  of  paQent  due  to  paQent   data  mix-­‐up   •  Body  part  damage  (e.g.  eye  or  fingers) 3 Moderate Results  in  temporary/  reversible  injury     or     temporary/reversible  impairment    requiring   professional  medical  or  surgical  intervenQon •  Incorrect  or  inadequate  medical  treatment           2 Minor Results  in  temporary  injury  or  impairment  not   requiring  professional  medical  intervenQon •  Minor  incorrect  or  inadequate  medical  treatment   •  Delayed  medical  treatment,    loss  of  sample/new  sample   required  or  no  result   •  Equipment  damage     •  Privacy  violaQon  (e.g.  data  leak) 1 Negligible Inconvenience  or  temporary  discomfort •  UnsaQsfied  user   •  Loss  of  old  data QuanQfying  Severity  &  Probability   •  Amount  of  severity  should  happen  by  considering   the  different  viewpoints  of  the  system’s   stakeholders.   •  The  probability  of  impact  can  only  happen   indirectly,  e.g.  by  evaluaQon  of     –  frequent  of  use,     –  quality  indicators  like  the  complexity  of  the  soeware   itself,     –  the  quality  of  the  documentaQon     –  etc.  

07/03/14   11   QuanQfying  Severity  &  Probability   •  Do  not  spend  too  much  Qme  determining  the   exact  values   •  The  relaQve  scoring  is  the  most  important  in   order  to  idenQfy  the  most  criQcal  parts.   •  The  scoring  could  be  rather  informal  and   based  on  a  brainstorm     Note:  For  Medical  Device  Class  B  and  Class  C  it   is  expected  that  the  Risk  Analysis  is  more   elaborated  than  stated  here   Risk  levels   •  Four  levels  (3  should  be  OK)   •  Based  on  IEC  62304  safety  class  (A,B,C)   •  ClassificaQon  used  for:   – Requirements  (i.e.  some  funcQons  are   more  criQcal,  e.g.  Risk  Control  Measures   than  others)   – Structural  elements  both  design  and   actual  implementaQon  (i.e.  elements   implemenQng  criQcal  requirements)   Least  criQcal   Most  criQcal   C2   B   A   C1  

07/03/14   12   Ploxng  the  risks   Severity   High  Low   Probability  High  Low   High  Risks   Medium  Risks   Low  Risks   C2   C1   B   A   Different  risk  perspecQves   For  a  Medical  Device:   •  Safety:  Freedom  of  unacceptable  risks.  IdenQfying  and  miQgaQng  safety   risks  to  paQents  and  users.     •  Effec@veness:  Fulfil  the  medical  claims,  delivering  value  to  the  paQent  and   users.  MeeQng  the  user’s  needs.  Correctness  of  the  product,  meeQng  its   specificaQons.   •  Customer  sa@sfac@on:  Good  user  experience,  good  service,  ease  of  use,   free  of  defects,  reliable.   •  Regulatory:  Being  in  compliance  by  meeQng  the  Regulatory  ExpectaQons   including:  Safety,  Efficacy  and  Customer  saQsfacQon.     •  Project:  MeeQng  the  organisaQons  expectaQons,  including  Quality,  Safety,   EffecQveness  and  Regulatory,  but  also  Qmelines  and  other  tradiQonal   project  risk  related  stuff.  

07/03/14   13   Different  risk  perspecQves   Safety   EffecQveness   Regulatory   Customer   SaQsfacQon   Project  risk   Regulatory  Risk  –  Process  Rigor  

07/03/14   14   Process  Rigor   Common  FDA  Warning  lefer  issues   – Lack  of     •  test  specificaQons  and  test  results   •  comprehensive,  up-­‐to-­‐date  specificaQons  (design  input)   – Inadequate     •  fault  handling  and  stress  tesQng   •  change  and  release  control   Regulatory  Risks  versus     ProducQvity  &  Predictability   •  RegulaQons  and  standards  does  not  seek  benefits   in  producQvity  or  project  predictability     •  But  they  don’t  preclude  producQvity  and   predictability  from  being  important.   •  So  it  is  your  responsibility  and  interest  to  have   development  processes  focusing  on:   –  ProducQvity   –  Predictability   Without  sacrificing  Safety,  Customer  saQsfacQon,   EffecQveness  and  Regulatory  risks  

07/03/14   15   Regulatory  Risks  versus     ProducQvity  &  Predictability   •  Risk  Driven  Approach:   –  Regulatory  interpretaQon  –  focus  on  the  intenQon   –  ConQnuously  process  improvement  within  the  intenQon  and  frame  of   the  Regulatory  expectaQons   –  Align  process  with  the  different  risks  (e.g.  Safety,  EffecQveness,  and   Customer  saQsfacQon)  associated  -­‐  focus  on  what  really  mafers   •  “Too  much  will  always  be  too  much,  but  maybe  not  enough”   Regulatory  Risk:  When  not  in  compliance   •  Compliance  is  not  created  by:   – using  a  checklist   – copying  from  the:   •  standards   •  regulaQons   •  guidance's   •  CreaQng  compliance  is  about  meeQng  the  “intent”  of  the   standards  –  not  just  following  “the  lefer  of  the  law”  

07/03/14   16   Regulatory  Risk  –  Don’t  climb  too  high   E.g.  Tools  validaQon  -­‐  Balancing  between:     •  what  is  required  and     •  when  value  adding  stops   minimum   opQmum   Risk  Based  DocumentaQon  Rigor  

07/03/14   17   DocumentaQon  Rigor   •  To  lifle   –  Difficult  to  anchor  decisions   –  High  regulatory  risk   –  Project  delay   –  Maintenance  is  difficult   •  To  much   –  AddiQonal  cost   –  Less  Qme  for  development  (project  delay)   –  Maintenance  is  difficult   –  Risk  of  in-­‐consistence   –  Project  delay   Requirements  Rigor   Rigor of requirementsHigh Low Safety UI User Satisfaction Service Risk Low High Efficacy

07/03/14   18   Design  Rigor   •  62304  allows  soeware  to  be   decomposed  into  soeware  items   with  different  safety  classes     – if  they  are  segregated  and  segregaQon   raQonale  provided   – No  raQonale  =>  Safety  Class  is  the   same  for  all  Item/Units.   – type  of  segregaQon  could  vary  based   on  risk  and  other  factors   Design  Rigor  –  Item/Unit  Level   Higher  Risk  Level  requires,   more  detailed  and  rigorous:     •  Architecture  descripQon   •  Detailed  Design     If  the  enQre  SW  is  Safety  Class   C  Detailed  design  is  required,   but  it  is  sQll  possible  to  use  the   Risk  Based  Approach  and   adjust  on  details  and  rigor       Least  criQcal   Component   Most  criQcal   Component  UI   FuncQon   Model   Driver   HAL   SI   C2   B   C2   B   A  A   C1   C1   Allocate  Risk  Level  to  the  different  Items/Units   based  on  their  responsibility  

07/03/14   19   Risk  Based  TesQng   Risk  Based  TesQng   •  IdenQfy  the  top  most  criQcal  funcQons     •  Consider:   –  Evaluate  whether  the  users  will  idenQfy  defects  in   funcQon  or  afribute.   –  Use  historical  data  to  idenQfy  funcQon  areas  with   many  defects   •  Do  extra  tesQng  in  criQcal  areas  and  areas  with   many  defects   –  Use  domain  specialists   –  Extend  (automated)  regression  test  when  new  defects   are  found  

07/03/14   20   For  System  TesQng   Risk  level   SW  System  test  ac@vi@es   C2,  C1   •  FuncQonal   •  Exploratory   •  Consider  other  test  strategies  (Stress,  Boundary,  Stability,  State   transsion,  Recovery)  dependent  of  the  FuncQon  under  test  and  Risk   level   B   •  FuncQonal   •  Security   •  Exploratory   A   •  FuncQonal  (all  requirements  have  at  least  one  TC)   •  Allocate  Risk  Level  to  the  different  FuncQons/ requirements  based  on  the  possible  severity  and   probability.   Item/Unit  Level  tesQng   Higher  Risk  Level  requires,  more   detailed  and  rigorous  QA   AcQviQes:   •  Reviews,   •  TesQng,   •  Etc.     If  appropriate  use  historical  data   to  idenQfy  Item/Units  with  many   defects  in  order  to  adjust  the  Risk   Level       Least  criQcal   Component   Most  criQcal   Component  UI   FuncQon   Model   Driver   HAL   SI   C2   B   C2   B   A  A   C1   C1  

07/03/14   21   Item/Unit  Level  tesQng   Risk  level   Unit  Tes@ng  ac@vi@es   C2   •  Formal  Code  Review  by  at  least  one  SW  Developer  plus  SW  Risk  Manager   •  StaQc  Analysis   •  Soeware  Unit  Test,  100%  decision/condiQon  coverage   •  IntegraQon  test  using  decisions  tables  and  classificaQon  trees   C1   •  Formal  Code  Review  by  at  least  one  SW  Developer   •  StaQc  Analysis     •  Soeware  Unit  Test,  100%  Statement  coverage   •  IntegraQon  test  using  decision  tables   B   •  Formal  Code  Review  by  at  least  one  SW  Developer   •  StaQc  Analysis   •  Soeware  Unit  Test,  100%  funcQon  coverage   •  IntegraQon  test     A   •  Informal  Peer  Code  Review   •  IntegraQon  test  part  of  System  Level  Test   Note:  Risk  Level  A  not  to  be  used  for  Class  C  Soeware   Item/Unit  Level  tesQng  &  Complexity   Risk   level   Complexity     Reduce   C2   McCabe  <=  2  AND  LoC  <  10   Reduce  Unit  Test  to  100%  FuncQon  coverage   C1   McCabe  <=  3  AND  LoC  <  20   Reduce  Unit  Test  to  100%  FuncQon  coverage   B   McCabe  <=  3  AND  LoC  <  30   No  Unit  Test  necessary  (Not  for  Class  C   Soeware   A   NA   NA   •  Complexity  –  root  cause  for  many  defects  –  but  low   complexity  code  may  also  require  less  tesQng.   •  In  source  code  use  complexity  metrics  to  adjust  the  QA   acQviQes   •  Note:  Complexity  metrics  also  part  of  the  code  standard  

07/03/14   22   ProacQve  QA   ProacQve  QA   •  Uses  same  approach  as  for  Medical  Device  Safety   Risk  Management  (IEC  14971,  IEC  80002).   •  IdenQfy  most  criQcal  SW  hazards  and  their   causes,  e.g.:   –  Loss  of  configuraQon  could  make  the  SW/Device   useless   –  Faulty  data  could  result  in  fault  funcQonality  and/or   results   –  Never-­‐ending  waiQng  loops  could  could  make  the  SW/ Device  slow  or  useless   •  Implement  proper  miQgaQons  in  the  design  

07/03/14   23   Examples   Example  –  Keyboard  in  ATM   Keyboard  Dispenser   Display   Card   Reader   Ext   Keyboard   XFS  drv   XFS  drv  XFS  drv   XFS   Win  XP   Bank  App   PC   ATM   Master-­‐key   derived-­‐key   derived-­‐key   derived-­‐key   derived-­‐key   •  <1%  source  code  doing   keyboard  funcQonality   •  Remaining  related  to:   •  Security:  crypt,  key-­‐ handling,  surveillance   •  Service:  ConfiguraQon,   status  log  etc.   Reliability  wise  –  key  handling  is  very  important   •  Without  the  Master  key  the  keyboard  needs  to   back  to  the  manufacturer   •  Without  derived  keys  the  keyboard  needs  a   service  tech.  visit    

07/03/14   24   Risk  Based  QA  for  ATM  keyboard   •  “Spontaneous”  loss  of  keys  in  the  field   •  Risk  based  approach:  New  file  system  with     – CRC  Error  correcQon     – “Black  box  recorder”  (log  of  field  events  for   debugging)   – More  rigor  of  requirements  and  design   – Unit  tesQng  of  the  new  file  system   Risk  Based  QA  for  ATM  keyboard   •  In  the  pilot  phase  -­‐  several  incidences  where   keyboard  is  “completely  dead”     •  In  certain  situaQons,  defects  in  both  the  CRC   Error  correcQon  and  “Black  box  recorder”  ends   up  in:  “logging  an  error  result  in  a  new  error…”   •  Learnings:   –  Adding  “miQgaQons”  in  soeware  increases  complexity   –  which  may  end  up  in  more  erroneous  SW   –  Remember  integraQon  and  scenario  tesQng     –  Consider  some  kind  of  recovery  mechanism  

07/03/14   25   Example  ProacQve  QA  for  VHF  radio   •  VHF  Radio  stores  “vital”  data  in  EEPROM.   •  During  SW  test  a  HW  design  flaw  is  found.  HW   do  not  give  “Power  Down”  in  Qme  =>  “vital”   data  is  corrupted  =>  VHF  Radio  is  useless.   •  ProacQve  QA:   – CRC  protecQon  of  data   – Shadowing  of  data   – Controlled  Scheme  for  data  update   – The  approach  also  miQgates  SW  failures   Bafery  monitor  in  Medical  Device   •  A  Bafery  powered  Medical  Device  have  a  bafery   monitor  to  inform  when  charging  is  needed   •  ProacQve  QA   –  Bafery  power  is  (also)  displayed  in  number  of   measurements  lee   –  If  number  of  measurements  lee  <=  2  then   measurements  is  not  possible   –  When  number  of  measurements  lee  <=2  then  it  is   always  decreased  with  1  independently  of  bafery   status   –  Monitoring  algorithm  adapts  when  bafery  degrades  

07/03/14   26   ICU  Monitor  in  Demo  mode   •  ICU  Monitor  have  a  demo  mode  to  show   realisQc  waveforms  in  sales  situaQon.   •  Erroneously  a  ICU  Monitor  jumped  into  Demo   mode  during  monitoring  of  real  paQent   •  ProacQve  QA   – Changing  waveform  to  non  realisQc  waveforms   – WriQng  “Demo”  where  waveforms  are  displayed   – Timeout  on  Demo  mode  –  jumping  back  to  real   mode  aeer  a  period  of  in-­‐acQvity   PiZalls  

07/03/14   27   PiZalls   •  Too  much  are  considered  criQcal  or  too  much   are  considered  not-­‐criQcal   – Risk  of  not  finding  the  criQcal  defects   •  Customer  and  Manufacturer  have  different   view  of  what  is  criQcal     – Risk  of  un-­‐saQsfied  customer   •  Management  only  buys  the  cost  reducQon   part  of  risk  based  QA   – Risk  of  poor  quality     PiZalls   •  No  use  of  historical  data  –  also  within  the   project   – If  defect  trends  shows  different  than  your  risk   evaluaQon  (e.g.  un-­‐idenQfied  criQcal  defects)  then   you  should  adapt  your  Risk  Based  Approach.   •  Design  miQgaQons  adds  too  much  complexity   – resulQng  in  other  defects,  difficult  to  maintain  SW   •  The  “system”  to  handle  Risk  Based  QA  are  too   complex.  

Add a comment

Related presentations

Presentación que realice en el Evento Nacional de Gobierno Abierto, realizado los ...

In this presentation we will describe our experience developing with a highly dyna...

Presentation to the LITA Forum 7th November 2014 Albuquerque, NM

Un recorrido por los cambios que nos generará el wearabletech en el futuro

Um paralelo entre as novidades & mercado em Wearable Computing e Tecnologias Assis...

Microsoft finally joins the smartwatch and fitness tracker game by introducing the...

Related pages

Michael Agerkvist Petersen | LinkedIn

... helping professionals like Michael Agerkvist Petersen discover inside ... Join LinkedIn and access Michael Agerkvist’s ... Radiometer Medical. QA ...
Read more

Lars Bo Klitgaard | LinkedIn

Som medlem af LinkedIn kommer du i selskab med 400 ... risk management ... Michael Agerkvist Petersen. Expert within Medical Device Software ...
Read more

Jens Peter Andersen | LinkedIn

View Jens Peter Andersen ... Introduction of risk management in Pallas Informatik based on ISO31000:2009 Risk ... Application of risk management to medical ...
Read more

Pairwise Testing - DevelopSense - Things can be different

This is called pairwise testing or ... One workaround is to complement pairwise testing with other testing and QA ... risk-based design in which ...
Read more

Home - DSM

Royal DSM is a global science based company active in health nutrition and materials. ... Risk management. System. ... Medical; Packaging & Graphic ...
Read more

Baylor College of Medicine | Houston, Texas

Baylor College of Medicine in Houston is recognized as a premier academic health science center and is known for excellence ... Find the best medical care.
Read more


Advertising Programmes Business Solutions +Google About Google Google.com © 2016 - Privacy - Terms ...
Read more