Upgrade to Pro — share decks privately, control downloads, hide ads and more …

AppDynamics Unified Monitoring

AppDynamics Unified Monitoring

AppDynamics Unified Monitoring presented at Velocity 2015


Dustin Whittle

May 28, 2015


  1. Unified  Monitoring  with   AppDynamics Dus$n  Whi*le   @AppDynamics

  2. None
  3. 52% of Fortune 500 firms since 2000 are gone

  4. Login Flight Status Search Flight Purchase Mobile Big data SOA

    NOSQL Cloud Agile Web Application complexity is exploding
  5. None
  6. None
  7. Up7me  is  cri7cal  for  enterprises  and  consumers

  8. Performance  impacts  the  bo=om  line

  9. How  fast  is  fast  enough? § Performance  is  key  to

     a  great  user  experience     -­‐ Under  100ms  is  perceived  as  reac$ng  instantaneously   -­‐ A  100ms  to  300ms  delay  is  percep$ble   -­‐ 1  second  is  about  the  limit  for  the  user's  flow  of  thought  to  stay  uninterrupted   -­‐ Users  expect  a  site  to  load  in  2  seconds   -­‐ ADer  3  seconds,  40%  will  abandon  your  site.   -­‐ 10  seconds  is  about  the  limit  for  keeping  the  user's  a*en$on   § Modern  applica7ons  spend  more  7me  in  the  browser  than  on  the  server-­‐side
  10. Who  cares  about  performance?

  11. None
  12. None
  13. None
  14. How  many  enterprise  monitoring  products  would  you  es7mate  your  IT

     org  owns?
  15. The  war  room  response  team

  16. The  problems  with  monitoring  tools § Root  cause  isola7on  is

     elusive  as  monitoring  lives  in  silos   -­‐ Infrastructure   -­‐ Hardware  +  Logs  +  Network  +  Storage  +  Containers  +  VMs   -­‐ Applica$on   -­‐ Load  Balancers  +  Web  Servers  +  App  Servers  +  File  Servers   -­‐ Databases  +  Caches  +  Queues  +  Third  party  services   -­‐ End  Users   -­‐ CDN   -­‐ Web  +  Mobile   § Metrics  lack  the  context  of  impact
  17. Monitoring  lacks  the  business  context

  18. The  struggle  of  modern  monitoring § Organiza7ons  focus  on  availability

     +  raw  metrics  and  not  end  user  experience  /  impact   § Complex  apps  built  on  micro-­‐services  in  containers  living  in  elas7c  cloud  environments   § Too  many  graphs  from  too  many  metrics   -­‐ Understanding  the  signal  from  the  noise  is  difficult   -­‐ No  topology  awareness.  No  transac$onal  visibility.  No  root  cause.   § Alert  storming  with  too  many  false  alarms   -­‐ Aler$ng  is  based  on  sta$c  thresholds  —  lacks  intelligent  anomaly  detec$on  +  correla$on   -­‐ Lack  of  historical  context  or  rela$onships  between  metrics  and  events   § A  single  pane  of  glass  across  the  performance  stakeholders   -­‐ Not  able  to  quan$fy  the  impact  of  performance  degrada$on  —  Not  self-­‐service
  19. None
  20. Context  is  king:  Unified  Monitoring

  21. Breaking  down  the  silos

  22. None
  23. None
  24. None
  25. Situa7on-­‐aware  data  and  views Web Ops App Owner Server Admin

    DBA IT Ops
  26. None
  27. None
  28. None
  29. None
  30. None
  31. Monitor  the  end  user  experience § Real  User  Monitoring  vs

     Synthe7c  Monitoring   -­‐ Synthe$c  tests  provide  24/7  assurance   -­‐ RUM  provides  insights  into  actual  users   -­‐ Mobile  device  segmenta$on   -­‐ Unexpected  behavior/trends   § Real  User  Monitoring   -­‐ Naviga$on  Timing  API   -­‐ Resource  Timing  API   -­‐ User  Timing  API   -­‐ Javascript  Errors
  32. Metrics  +  logs  help,  but  intelligence  is  be=er

  33. None
  34. None
  35. None
  36. Moving  from  reac7ve  to  proac7ve § Resolving  before  the  red

     =  fixing  in  the  yellow   -­‐ Automa$c  runbook  automa$on  integrates  with  your  devops  stack   § Intelligent  anomaly  detec7on  across  end-­‐user,  applica7on,  database,  server  metrics   -­‐ Automa$cally  calculates  dynamic  baselines  for  all  of  your  metrics,  which,  based  on  actual   usage,  define  what  is  "normal"  for  each  metric   -­‐ Smart  aler$ng  based  on  any  devia$on  from  the  baselines   § Understand  trends  and  pa=erns  in  failures  -­‐  automa7cally  learn  from  the  past   -­‐ Understand  what  are  the  most  impacaul  issues  to  resolve   -­‐ Ocen  $mes  external  services  are  the  root  cause  with  limited  visibility   -­‐ Enforce  SLAs  
  37. None
  38. Moving  from  reac7ve  to  proac7ve -­‐ Automa7c  discovery  of  environment

     and  applica7on  changes   -­‐ New  APIs,  transac$ons,  services,  clouds   § Leverage  analy7cs  to  be  smarter  about  using  the  data  you  already  have   -­‐ System  Logs,  Metrics  from  events  and  infrastructure  stats   -­‐ Transac$ons  with  request  parameters  +  User  state  from  cookies/sessions   § Performance  monitoring  isn’t  just  about  the  tech   -­‐ Visibility  into  the  impact  of  business  -­‐  aler$ng  when  revenue  is  down
  39. AppDynamics  leverages  and  embraces  open-­‐source

  40. Leading  companies  invest  in  performance § Etsy  =  Kale  =

     Statsd  +  Skyline  +  Oculus  (stats  collec$on  +  anomaly  detec$on/correla$on)   § Nealix  =  PCP  +  Vector  +  Servo  +  Atlas  (dashboards,  data  collec$on,  root  cause  analysis)   § Twi*er  =  Zipkin  (distributed  tracing)
  41. Recommenda7ons § Treat  performance  as  a  feature     -­‐

    Create  a  performance  budget  with  milestones,  speed  index,  page  speed   -­‐ Capacity  plan  and  load  test  the  server-­‐side     -­‐ Op$mize  and  performance  test  the  client-­‐side     § Monitor  performance  in  development  and  produc7on     -­‐ Instrument  everything   -­‐ Measure  the  difference  of  every  change   -­‐ Understand  how  failures  impact  performance   § Make  monitoring  cri7cal  and  test  in  your  con7nuous  delivery  process     § Connect  the  exec/dev/ops  performance  perspec7ves  to  align  on  business  impact
  42. Go  back  and  inves7gate  how  your  company   can  break

     down  the  monitoring  silos  and  be   more  impac]ul  with  applica7on  intelligence.
  43. Ques7ons?

  44. Thank  you!  Enjoy  the  rest  of  Velocity  2015.

  45. http://www.appdynamics.com/