• Share
  • Email
  • Embed
  • Like
  • Save
  • Private Content
Keynote: JP Schmetz - How i learned to stop worrying and love
 

Keynote: JP Schmetz - How i learned to stop worrying and love

on

  • 664 views

Jean-Paul Schmetz, HackFwD - How I learned to stop worrying and love the Cloud ...

Jean-Paul Schmetz, HackFwD - How I learned to stop worrying and love the Cloud
---
Please contact us for a downloadable copy of the slides at CloudOps.Summit@googlemail.com .

Follow us on Twitter @CloudOps_Summit and
Facebook http://www.facebook.com/CloudOps

Statistics

Views

Total Views
664
Views on SlideShare
664
Embed Views
0

Actions

Likes
0
Downloads
0
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    Keynote: JP Schmetz - How i learned to stop worrying and love Keynote: JP Schmetz - How i learned to stop worrying and love Presentation Transcript

    • How  I  learned  to  stop  worrying   and  love  the  cloud   Cloudops     17-­‐03-­‐2011  
    • An  Epiphany  389  
    • An  Epiphany  389   Boulder  Creek  Fault  
    • An  Epiphany  389   389   Boulder  Creek  Fault  
    • An  Epiphany  389   389   Boulder  Creek  Fault   So  I  asked  for  more  informaGon…  
    • And  the  first  few  pages  of  the  presentaGon     were  pictures  of  machines  
    • To  me  it  looked  like  this  And  I  decided  to  understand  what  was  wrong  
    • What’s  wrong  with  Hardware  1.  It’s  a  fixed  asset  that  needs  to  be  planned  
    • What’s  wrong  with  Hardware  1.  It’s  a  fixed  asset  that  needs  to  be  planned   –  In  most  cases,  you  cannot  plan  for  any  Gmeframe  
    • What’s  wrong  with  Hardware  1.  It’s  a  fixed  asset  that  needs  to  be  planned   –  In  most  cases,  you  cannot  plan  for  any  Gmeframe   –  So  people  overplan  (too  many  machines,  too   many  top-­‐line  configuraGons)  
    • What’s  wrong  with  Hardware  1.  It’s  a  fixed  asset  that  needs  to  be  planned   –  In  most  cases,  you  cannot  plan  for  any  Gmeframe   –  So  people  overplan  (too  many  machines,  too   many  top-­‐line  configuraGons)   –  And  you  pay  (or  commit  to  pay)  in  advance.  Bite-­‐ size  increases  difficult  because  of  approval  
    • What’s  wrong  with  Hardware  2.  Once  the  machines  are  there  
    • What’s  wrong  with  Hardware  2.  Once  the  machines  are  there   –  ApplicaGons  and  data  will  fill  them  like  gas  
    • What’s  wrong  with  Hardware  2.  Once  the  machines  are  there   –  ApplicaGons  and  data  will  fill  them  like  gas   –  If  your  requirements  change,  machines  are   simply  unused  
    • What’s  wrong  with  Hardware  2.  Once  the  machines  are  there   –  ApplicaGons  and  data  will  fill  them  like  gas   –  If  your  requirements  change,  machines  are   simply  unused   –  You  cannot  scale  up  fast  and  you  certainly   cannot  scale  down  at  any  speed  
    • What’s  wrong  with  Hardware  3.  Once  a  machine  is  installed  
    • What’s  wrong  with  Hardware  3.  Once  a  machine  is  installed   –  You  program  and  configure  it  as  if  it  will  remain   the  same  forever  
    • What’s  wrong  with  Hardware  3.  Once  a  machine  is  installed   –  You  program  and  configure  it  as  if  it  will  remain   the  same  forever   –  SoXware  and  hardware  become  too   interconnected  someGmes  even  by  design  
    • What’s  wrong  with  Hardware  4.  There  is  NEVER  an  opGmal  to  think  or  act   differently  
    • What’s  wrong  with  Hardware  4.  There  is  NEVER  an  opGmal  to  think  or  act   differently   –  If  you  introduce  the  idea  of  cloud,  people  will   approach  as  if  it  were  alternaGve  hosGng  and   your  vendor  will  gladly  underprice  to  keep  you   hooked  
    • What’s  wrong  with  Hardware  4.  There  is  NEVER  an  opGmal  to  think  or  act   differently   –  If  you  introduce  the  idea  of  cloud,  people  will   approach  as  if  it  were  alternaGve  hosGng  and   your  vendor  will  gladly  underprice  to  keep  you   hooked   –  Your  applicaGon  developers  will  predict  mayhem  
    • What’s  wrong  with  Hardware  4.  There  is  NEVER  an  opGmal  to  think  or  act   differently   –  If  you  introduce  the  idea  of  cloud,  people  will   approach  as  if  it  were  alternaGve  hosGng  and   your  vendor  will  gladly  underprice  to  keep  you   hooked   –  Your  applicaGon  developers  will  predict  mayhem   –  As  a  compromise,  your  tech  people  (you?)  will   propose  to  build  an  internal  cloud  system…  
    • It’s  not  an  evoluGon  
    • It’s  not  an  evoluGon  •  Using  the  cloud  like  you  use  a  fixed  asset   system  will  never  yield  any  advantage  
    • It’s  not  an  evoluGon  •  Using  the  cloud  like  you  use  a  fixed  asset   system  will  never  yield  any  advantage  •  You  have  to  think  and  build  differently  
    • It’s  not  an  evoluGon  •  Using  the  cloud  like  you  use  a  fixed  asset   system  will  never  yield  any  advantage  •  You  have  to  think  and  build  differently  •  Learn  from  people  who  can  afford  to  start   from  scratch  (start-­‐ups)  
    • How  to  think  right  •  Hardware  is  soXware  (it  always  sort  of  was…)  
    • How  to  think  right  •  Hardware  is  soXware  (it  always  sort  of  was…)   cpu  
    • How  to  think  right  •  Hardware  is  soXware  (it  always  sort  of  was…)   cpu   ec2-­‐run-­‐instances  ami-­‐6138dd08  -­‐k  pstam-­‐keypair      
    • How  to  think  right  •  Hardware  is  soXware  (it  always  sort  of  was…)   ram   cpu   ec2-­‐run-­‐instances  ami-­‐6138dd08  -­‐k  pstam-­‐keypair      
    • How  to  think  right  •  Hardware  is  soXware  (it  always  sort  of  was…)  ec2-­‐create-­‐volume  –size  10   FS   ram   cpu  
    • How  to  think  right  •  Hardware  is  soXware  (it  always  sort  of  was…)   FS   ram  ec2-­‐acach-­‐volume  volume_id  -­‐-­‐instance  instance_id  -­‐-­‐device  device     cpu  
    • How  to  think  right  •  Hardware  is  soXware  (it  always  sort  of  was…)   FS   ram   queues   storage   DNS/ email   Load   cpu  
    • What  do  you  get?  
    • What  do  you  get?  •  It’s  basically  an  object  or  a  class.  
    • What  do  you  get?  •  It’s  basically  an  object  or  a  class.   –  You  can  construct  it  
    • What  do  you  get?  •  It’s  basically  an  object  or  a  class.   –  You  can  construct  it   –  It  can  do  inheritance/delegaGon  (get  the  code  to   do  your  work  from  a  repository,  assume  a  role)  
    • What  do  you  get?  •  It’s  basically  an  object  or  a  class.   –  You  can  construct  it   –  It  can  do  inheritance/delegaGon  (get  the  code  to   do  your  work  from  a  repository,  assume  a  role)   –  Get  the  data  it  needs  (from  scratch  or  cloning  it   from  another  object)  
    • What  do  you  get?  •  It’s  basically  an  object  or  a  class.   –  You  can  construct  it   –  It  can  do  inheritance/delegaGon  (get  the  code  to   do  your  work  from  a  repository,  assume  a  role)   –  Get  the  data  it  needs  (from  scratch  or  cloning  it   from  another  object)   –  ParGcipate  in  the  system  unGl  told  to  stop  or  it   crashes  
    • What  do  you  get?  •  It’s  basically  an  object  or  a  class.   –  You  can  construct  it   –  It  can  do  inheritance/delegaGon  (get  the  code  to   do  your  work  from  a  repository,  assume  a  role)   –  Get  the  data  it  needs  (from  scratch  or  cloning  it   from  another  object)   –  ParGcipate  in  the  system  unGl  told  to  stop  or  it   crashes   –  Combine  to  create  more  complex  objects  
    • Example  Search   Search  Search   Search   Search  object   object   object  object   object   1000x  
    • Example   input   sns  Search   Search  Search   Search   Search  object   object   object  object   object   1000x  
    • Example   input   sns  Search   Search  Search   Search   Search  object   object   object  object   object   1000x   load  balancer   End  user  
    • How  to  upgrade?   input   sns  Search   Search  Search   Search   Search  object   object   object  object   object   1000x   load  balancer   End  user  
    • How  to  upgrade?   input   In  a  fixed    asset  environment   there  are  two  possibiliGes:   sns    Search   Search  Search   Search   Search  object   object   object  object   object   1000x   load  balancer   End  user  
    • How  to  upgrade?   input   In  a  fixed    asset  environment   there  are  two  possibiliGes:   sns     Overplan  mode:    Search   Search  Search   Search   Search   -­‐  Upgrade  the  part  that  is  object   object   object  object   object   1000x   unused  offline   -­‐  Switch  between  versions  on   the  front-­‐end   load  balancer     End  user  
    • How  to  upgrade?   input   In  a  fixed    asset  environment   there  are  two  possibiliGes:   sns     Overplan  mode:    Search   Search  Search   Search   Search   -­‐  Upgrade  the  part  that  is  object   object   object  object   object   1000x   unused  offline   -­‐  Switch  between  versions  on   the  front-­‐end   load  balancer   Full  capacity:     -­‐  Use  the  maintenance   window  and  pray  that  it   works  out   End  user    
    • How  to  upgrade?   input   sns   In  any  case:  it’s  a  mess  +  if  not  Search   Search  Search   Search   Search  object   object   object  object   object   automated,  lots  of  manual  sys  admin   1000x   work   load  balancer   End  user  
    • How  to  upgrade?   input   sns  Search   Search  Search   Search   Search  object   object   object  object   object   dev   1000x   load  balancer   End  user  
    • How  to  upgrade?   input   sns   sns2  Search   Search  Search   Search   Search  object   object   object  object   object   dev   1000x   load  balancer   End  user  
    • How  to  upgrade?   input   sns   sns2  Search   Search  Search   Search   Search   Search   Search  Search   Search   Search  object   object   object  object   object   dev   object   object   object   object  object   1000x   1000x   load  balancer   End  user  
    • How  to  upgrade?   input   sns   sns2  Search   Search  Search   Search   Search   Search   Search  Search   Search   Search  object   object   object  object   object   dev   object   object   object   object  object   1000x   1000x   load  balancer   End  user  
    • How  to  upgrade?  input   sns2   Search   Search  Search   Search   Search   dev   object   object   object   object  object   1000x   load  balancer   End  user  
    • How  to  upgrade?   input   sns2  Total  Cost:     Search   Search  Search   Search   Search   dev   object   object   object   object  object  -­‐  Duplicate  system  for   1000x   a  few  hours  ($170/ hour)  -­‐  No  manual  labor  –   cloning  is  just  a  script  -­‐  No  risk  of  failure   load  balancer  -­‐  No  maintenance   window   End  user  
    • How  to  downgrade?  input   sns2   Search   Search  Search   Search   Search   dev   object   object   object   object  object   1000x   load  balancer   End  user  
    • How  to  downgrade?   input   sns2  Total  Cost:     Search   Search  Search   Search   Search   dev   object   object   object   object  object  -­‐  500  ec2-­‐stop-­‐ 500x   instances  calls  -­‐  Maybe  calls  to  load   balancer  -­‐  Immediate  results   load  balancer   End  user  
    • FAQ  •  Costs  –  do  you  save  money?  
    • FAQ  •  Costs  –  do  you  save  money?   –  Not  so  much  if  you  do  not  program  differently    
    • FAQ  •  Costs  –  do  you  save  money?   –  Not  so  much  if  you  do  not  program  differently   –  Massively  if  you  do    
    • FAQ  •  Costs  –  do  you  save  money?   –  Not  so  much  if  you  do  not  program  differently   –  Massively  if  you  do   •  Always  right  size  –  immediate  benefits    
    • FAQ  •  Costs  –  do  you  save  money?   –  Not  so  much  if  you  do  not  program  differently   –  Massively  if  you  do   •  Always  right  size  –  immediate  benefits   •  Never  unused  capacity  because  of  mismatches    
    • FAQ  •  Costs  –  do  you  save  money?   –  Not  so  much  if  you  do  not  program  differently   –  Massively  if  you  do   •  Always  right  size  –  immediate  benefits   •  Never  unused  capacity  because  of  mismatches   •  Never  stuck  in  dead  ends  where  you  cannot  upgrade    
    • FAQ  •  Reliability:   –  Cloud  service  are  not  less  nor  more  reliable  than   other  systems.    
    • FAQ  •  Reliability:   –  Cloud  service  are  not  less  nor  more  reliable  than   other  systems.     –  There  is  an  ownership  bias  at  work  though  
    • FAQ  •  Reliability:   –  Cloud  service  are  not  less  nor  more  reliable  than   other  systems.     –  There  is  an  ownership  bias  at  work  though  •  Reliability  through  soXware  
    • FAQ  •  Reliability:   –  Cloud  service  are  not  less  nor  more  reliable  than   other  systems.     –  There  is  an  ownership  bias  at  work  though  •  Reliability  through  soXware   –  Instances  can  take  themselves  out  of  loop  when   they  have  trouble  
    • FAQ  •  Reliability:   –  Cloud  service  are  not  less  nor  more  reliable  than   other  systems.     –  There  is  an  ownership  bias  at  work  though  •  Reliability  through  soXware   –  Instances  can  take  themselves  out  of  loop  when   they  have  trouble   –  Use  a  chaos  monkey  to  make  sure  your  system  is   safe    
    • FAQ  •  Reliability:   –  Cloud  service  are  not  less  nor  more  reliable  than   other  systems.     –  There  is  an  ownership  bias  at  work  though  •  Reliability  through  soXware   –  Instances  can  take  themselves  out  of  loop  when  they   have  trouble   –  Use  a  chaos  monkey  to  make  sure  your  system  is  safe   –  Assume  failure  in  the  code  
    • FAQ  •  Clusterf**k  
    • FAQ  •  Clusterf**k   –  One  Amazon  data  center  goes  down  completely  
    • FAQ  •  Clusterf**k   –  One  Amazon  data  center  goes  down  completely   –  Not  really  a  problem:  you  can  be  in  up  to  7  data   centers  at  the  same  Gme.  
    • FAQ  •  Clusterf**k   –  One  Amazon  data  center  goes  down  completely   –  Not  really  a  problem:  you  can  be  in  up  to  7  data   centers  at  the  same  Gme.  •  Clusterf**k^2  
    • FAQ  •  Clusterf**k   –  One  Amazon  data  center  goes  down  completely   –  Not  really  a  problem:  you  can  be  in  up  to  7  data   centers  at  the  same  Gme.  •  Clusterf**k^2   –  All  Amazon  data  centers  go  down  at  the  same   Gme  and  all  the  data  is  lost  forever  
    • FAQ  •  Clusterf**k   –  One  Amazon  data  center  goes  down  completely   –  Not  really  a  problem:  you  can  be  in  up  to  7  data   centers  at  the  same  Gme.  •  Clusterf**k^2   –  All  Amazon  data  centers  go  down  at  the  same   Gme  and  all  the  data  is  lost  forever   –  Somewhat  a  problem,  but  all  your  system  images   and  volumes  can  be  saved  on  site  and  reloaded   elsewhere  (e.g.  with  Nimbula)  
    • FAQ  •  Data  ProtecGon  Laws  
    • FAQ  •  Data  ProtecGon  Laws   –  Usually  the  last  (desperate)  argument  
    • FAQ  •  Data  ProtecGon  Laws   –  Usually  the  last  (desperate)  argument   –  You  can  have  data  domesGcally  
    • FAQ  •  Data  ProtecGon  Laws   –  Usually  the  last  (desperate)  argument   –  You  can  have  data  domesGcally   –  It’s  Unix  –  encrypt  the  data  you  need  encrypted  
    • How  to  get  there  •  Start  with  what  you  build  yourself  and   experiment  
    • How  to  get  there  •  Start  with  what  you  build  yourself  and   experiment  •  Try  to  expose  yourself  to  startups  and  see   how  they  do  what  they  do  
    • How  to  get  there  •  Start  with  what  you  build  yourself  and   experiment  •  Try  to  expose  yourself  to  startups  and  see   how  they  do  what  they  do  •  If  you  use  vendors  system  and  just  install  and   maintain  them,  educate  yourself  and  try  to   find  a  vendor  that  thinks  right  
    • How  to  get  there  •  Start  with  what  you  build  yourself  and   experiment  •  Try  to  expose  yourself  to  startups  and  see   how  they  do  what  they  do  •  If  you  use  vendors  system  and  just  install  and   maintain  them,  educate  yourself  and  try  to   find  a  vendor  that  thinks  right  •  If  your  revenue/job  depends  on  the   inefficiency  being  solved,  start  to  pivot    
    • Thank  you     and  Good  Luck!