THE STATE OF GLOBAL INFRASTRUCTURE                  PERFORMANCE                        SPEAKER:   Marty Kagan             ...
The	  State	  of	  Global	  Infrastructure	  Performance                                                   Marty	  Kagan  ...
Community	  Measurements                        How	  to	  collect	  a	  billion	  measurements	  a	  dayFriday, July 27, ...
1.	  Publish	  content	  and	  applica3ons	  on	  200+	                             	  	  	  	  public	  and	  private	  c...
2.	  Deploy	  javascript	  tag	  on	  250+	  community-­‐member	  	  	                          	  	  	  	  websitesFriday...
3.	  Collect	  end-­‐user	  telemetry	  from	  30k	  networks	                          	  	  	  	  across	  200+	  countr...
4.	  Use	  the	  data	  to	  tell	  stories…	  Friday, July 27, 2012
4.	  Use	  the	  data	  to	  tell	  stories…	                                                                             ...
4.	  Use	  the	  data	  to	  tell	  stories…	                                                                             ...
4.	  Use	  the	  data	  to	  tell	  stories…	                                                                             ...
Story	  1:	  Cloud	  Availability                        How	  reliably	  reachable	  is	  Google	  App	  Engine	         ...
Google	  App	  Engine:	  20	  May	  2012                        90th	  Percen7le	  Response	  TimesFriday, July 27, 2012
Google	  App	  Engine:	  21	  May	  2012                        90th	  Percen7le	  Response	  TimesFriday, July 27, 2012
Google	  App	  Engine:	  22	  May	  2012                        90th	  Percen7le	  Response	  TimesFriday, July 27, 2012
Conclusion:             AVOID	  SINGLE-­‐VENDOR	  DEPENDENCIESFriday, July 27, 2012
Story	  2:	  Cloud	  Performance                        Why	  should	  I	  deploy	  my	  applica:ons	  across	  mul:ple	  ...
Amazon	  EC2	  US	  East	  (Virginia)                        Median	  Response	  TimesFriday, July 27, 2012
Amazon	  EC2	  EU	  West	  (Ireland)                        Median	  Response	  TimesFriday, July 27, 2012
Amazon	  EC2	  South	  America	  (Brazil)                        Median	  Response	  TimesFriday, July 27, 2012
Amazon	  EC2	  Asia	  Southeast	  (Singapore)                        Median	  Response	  TimesFriday, July 27, 2012
Amazon	  EC2	  (Performance	  Blend)                        Median	  Response	  TimesFriday, July 27, 2012
Windows	  Azure	  US	  South                        Median	  Response	  TimesFriday, July 27, 2012
Windows	  Azure	  EU	  North                        Median	  Response	  TimesFriday, July 27, 2012
Windows	  Azure	  Asia	  East                        Median	  Response	  TimesFriday, July 27, 2012
Windows	  Azure	  (Performance	  Blend)                        Median	  Response	  TimesFriday, July 27, 2012
VoxCloud	  Dallas                        Median	  Response	  TimesFriday, July 27, 2012
VoxCloud	  Amsterdam                        Median	  Response	  TimesFriday, July 27, 2012
VoxCloud	  Singapore                        Median	  Response	  TimesFriday, July 27, 2012
VoxCloud	  (Performance	  Blend)                        Median	  Response	  TimesFriday, July 27, 2012
Conclusion:                        REACH	  FURTHER	  WITH	  MULTI-­‐CLOUD	  STRATEGIESFriday, July 27, 2012
Story	  3:	  Blending	  clouds                        What	  combina:on	  of	  providers	  will	  deliver	  the	  best	  o...
1	  Clouds                                      !"#$%&()*                                        !"#$%&#                  ...
2	  Clouds                                      !"#$%&()*                                   +%,)-%./#                     ...
3	  Clouds                                      !"#$%&()*                                   +%,)-%./#     0#1234#15#      ...
4	  Clouds                                      !"#$%&()*                                   +%,)-%./#     0#1234#15#      ...
5	  Clouds                                      !"#$%&()*                                   +%,)-%./#     0#1234#15#      ...
Historical	  Latency-­‐Based	  Rou3ng                                    !"#$%&()*                                   +%,)-...
Historical	  Latency-­‐Based	  Rou3ng                                    !"#$%&()*                                   +%,)-...
Historical	  Latency-­‐Based	  Rou3ng                                    !"#$%&()*                                   +%,)-...
Historical	  Latency-­‐Based	  Rou3ng                                    !"#$%&()*                                   +%,)-...
Real-­‐3me	  Data-­‐Driven	  Rou3ng                                    !"#$%&()*                                         +...
Real-­‐3me	  Data-­‐Driven	  Rou3ng                                    !"#$%&()*                                         +...
Real-­‐3me	  Data-­‐Driven	  Rou3ng                                    !"#$%&()*                                         +...
Real-­‐3me	  Data-­‐Driven	  Rou3ng                                    !"#$%&()*                                         +...
!"#$%&()*                                     +%,)-%./#             0#1234#15#                 !$.65#                     ...
So,	  what	  have	  we	  learned	  today?                         %!"                         $!"                         ...
So,	  what	  have	  we	  learned	  today?                                                         •   Single-­‐plaRorm	  s...
So,	  what	  have	  we	  learned	  today?                                                         •   Single-­‐plaRorm	  s...
So,	  what	  have	  we	  learned	  today?                                                         •   Single-­‐plaRorm	  s...
What	  We	  Want                                To	  make	  the	  web	  faster	  for	                                   ev...
Friday, July 27, 2012
Upcoming SlideShare
Loading in...5
×

THE STATE OF GLOBAL INFRASTRUCTURE PERFORMANCE from Structure 2012

434

Published on

Presentation from Marty Kagan, Cedexis
#structureconf
More at http://event.gigaom.com/structure/

Published in: Technology, Business
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total Views
434
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

THE STATE OF GLOBAL INFRASTRUCTURE PERFORMANCE from Structure 2012

  1. 1. THE STATE OF GLOBAL INFRASTRUCTURE PERFORMANCE SPEAKER: Marty Kagan Co-Founder and CEO CedexisFriday, July 27, 2012
  2. 2. The  State  of  Global  Infrastructure  Performance Marty  Kagan marty@cedexis.comFriday, July 27, 2012
  3. 3. Community  Measurements How  to  collect  a  billion  measurements  a  dayFriday, July 27, 2012
  4. 4. 1.  Publish  content  and  applica3ons  on  200+          public  and  private  clouds Public  IaaS  &  PaaS Virtualized  Datacenters Global  &  Regional Pla,orms Delivery  NetworksFriday, July 27, 2012
  5. 5. 2.  Deploy  javascript  tag  on  250+  community-­‐member              websitesFriday, July 27, 2012
  6. 6. 3.  Collect  end-­‐user  telemetry  from  30k  networks          across  200+  countriesFriday, July 27, 2012
  7. 7. 4.  Use  the  data  to  tell  stories…  Friday, July 27, 2012
  8. 8. 4.  Use  the  data  to  tell  stories…   • How  reliable  is  a  single  Cloud  pla6orm?Friday, July 27, 2012
  9. 9. 4.  Use  the  data  to  tell  stories…   • How  reliable  is  a  single  Cloud  pla6orm? • Why  deploy  across  mul:ple  cloud  regions?Friday, July 27, 2012
  10. 10. 4.  Use  the  data  to  tell  stories…   • How  reliable  is  a  single  Cloud  pla6orm? • Why  deploy  across  mul:ple  cloud  regions? • What  combina:on  of  providers  will  deliver   the  best  performance?Friday, July 27, 2012
  11. 11. Story  1:  Cloud  Availability How  reliably  reachable  is  Google  App  Engine   from  countries/networks  around  the  world?Friday, July 27, 2012
  12. 12. Google  App  Engine:  20  May  2012 90th  Percen7le  Response  TimesFriday, July 27, 2012
  13. 13. Google  App  Engine:  21  May  2012 90th  Percen7le  Response  TimesFriday, July 27, 2012
  14. 14. Google  App  Engine:  22  May  2012 90th  Percen7le  Response  TimesFriday, July 27, 2012
  15. 15. Conclusion: AVOID  SINGLE-­‐VENDOR  DEPENDENCIESFriday, July 27, 2012
  16. 16. Story  2:  Cloud  Performance Why  should  I  deploy  my  applica:ons  across  mul:ple  Azure  or  EC2   regions?Friday, July 27, 2012
  17. 17. Amazon  EC2  US  East  (Virginia) Median  Response  TimesFriday, July 27, 2012
  18. 18. Amazon  EC2  EU  West  (Ireland) Median  Response  TimesFriday, July 27, 2012
  19. 19. Amazon  EC2  South  America  (Brazil) Median  Response  TimesFriday, July 27, 2012
  20. 20. Amazon  EC2  Asia  Southeast  (Singapore) Median  Response  TimesFriday, July 27, 2012
  21. 21. Amazon  EC2  (Performance  Blend) Median  Response  TimesFriday, July 27, 2012
  22. 22. Windows  Azure  US  South Median  Response  TimesFriday, July 27, 2012
  23. 23. Windows  Azure  EU  North Median  Response  TimesFriday, July 27, 2012
  24. 24. Windows  Azure  Asia  East Median  Response  TimesFriday, July 27, 2012
  25. 25. Windows  Azure  (Performance  Blend) Median  Response  TimesFriday, July 27, 2012
  26. 26. VoxCloud  Dallas Median  Response  TimesFriday, July 27, 2012
  27. 27. VoxCloud  Amsterdam Median  Response  TimesFriday, July 27, 2012
  28. 28. VoxCloud  Singapore Median  Response  TimesFriday, July 27, 2012
  29. 29. VoxCloud  (Performance  Blend) Median  Response  TimesFriday, July 27, 2012
  30. 30. Conclusion: REACH  FURTHER  WITH  MULTI-­‐CLOUD  STRATEGIESFriday, July 27, 2012
  31. 31. Story  3:  Blending  clouds What  combina:on  of  providers  will  deliver  the  best  overall   performance  in  the  United  States?Friday, July 27, 2012
  32. 32. 1  Clouds !"#$%&()* !"#$%&# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! +,-./0!+12340!Friday, July 27, 2012
  33. 33. 2  Clouds !"#$%&()* +%,)-%./# !"#$%&# ()*#+,-.# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.!Friday, July 27, 2012
  34. 34. 3  Clouds !"#$%&()* +%,)-%./# 0#1234#15# !"#$%&# ()*#+,-.# /0,12")-3)-&# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.!Friday, July 27, 2012
  35. 35. 4  Clouds !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.!Friday, July 27, 2012
  36. 36. 5  Clouds !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.!Friday, July 27, 2012
  37. 37. Historical  Latency-­‐Based  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>!Friday, July 27, 2012
  38. 38. Historical  Latency-­‐Based  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>!Friday, July 27, 2012
  39. 39. Historical  Latency-­‐Based  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>!Friday, July 27, 2012
  40. 40. Historical  Latency-­‐Based  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>!Friday, July 27, 2012
  41. 41. Real-­‐3me  Data-­‐Driven  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>! !+:78"=?:!@747"@51A:.!+,-=.>!Friday, July 27, 2012
  42. 42. Real-­‐3me  Data-­‐Driven  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>! !+:78"=?:!@747"@51A:.!+,-=.>!Friday, July 27, 2012
  43. 43. Real-­‐3me  Data-­‐Driven  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>! !+:78"=?:!@747"@51A:.!+,-=.>!Friday, July 27, 2012
  44. 44. Real-­‐3me  Data-­‐Driven  Rou3ng !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>! !+:78"=?:!@747"@51A:.!+,-=.>!Friday, July 27, 2012
  45. 45. !"#$%&()* +%,)-%./# 0#1234#15# !$.65# !$.65# !"#$%&# ()*#+,-.# /0,12")-3)-&# !"#(,-4# !"#",14# !(&$!! !(%$!! !(#$!! !"#$%&(")*+&)",$-".-)/ !($$!! !$!! !&$!! !%$!! !#$!! !"!!!! (! #! )! %! *! !+,-./!+,01.! !2134,51678!974:.6;!<73:/!+,-=.>! !+:78"=?:!@747"@51A:.!+,-=.>! Conclusion: NOT  ALL  BLENDS  ARE  CREATED  EQUALFriday, July 27, 2012
  46. 46. So,  what  have  we  learned  today? %!" $!" #!" !!" &!" %!" $!" #!" !" " #" (" $" )"Friday, July 27, 2012
  47. 47. So,  what  have  we  learned  today? • Single-­‐plaRorm  strategies  are  dangerous %!" $!" #!" !!" &!" %!" $!" #!" !" " #" (" $" )"Friday, July 27, 2012
  48. 48. So,  what  have  we  learned  today? • Single-­‐plaRorm  strategies  are  dangerous • Mul7-­‐cloud  strategies  are  key  to  reaching  a  global  audience %!" $!" #!" !!" &!" %!" $!" #!" !" " #" (" $" )"Friday, July 27, 2012
  49. 49. So,  what  have  we  learned  today? • Single-­‐plaRorm  strategies  are  dangerous • Mul7-­‐cloud  strategies  are  key  to  reaching  a  global  audience %!" $!" • Fresher  data  =  beUer  decisions #!" !!" &!" %!" $!" #!" !" " #" (" $" )"Friday, July 27, 2012
  50. 50. What  We  Want To  make  the  web  faster  for   every  user  on  the  planet.Friday, July 27, 2012
  51. 51. Friday, July 27, 2012

×