Making It Better Without Making It Over (Front Porch)

Transcript

1. Making It Better 
 Without Making It Over Rebecca Murphey

   @rmurphey Front Porch Dallas / October 2016
2. None
3. None
4. None
5. None
6. None

7. 2007

8. None
9. None
10. None
11. None
12. None
13. None
14. None
15. None
16. None
17. None
18. None
19. None
20. None
21. None
22. None

23. 0 5000 10000 15000 20000 2008 2009 2010 2011 2012

    2013 2014 2015 2016 2016.8 JS Lines of Code

24. Choose-your-
 own-adventure
 architecture

25. None

26. It’s easy to feel 
 like the only way 


    to make things better 
 is to start over.
27. None

28. a single, central source of state

29. None
30. None

31. 40 arguments

32. None

33. 2 arguments shared models

34. guardrails

35. None
36. None
37. None

38. crafting our deployment a new understanding of performance

39. None
40. None

41. Easy Math T = time S = payload size W

    = bandwidth (bytes/s) S = 78K W = 1500K/s T = S / W = 52ms

42. But latency … Data must travel from client to server

    and back.

43. Latency T = time S = payload size W =

    bandwidth (bytes/s) P = one-way transmission time (ping) S = 78K W = 1500K/s P = 75ms T = (2 * P) + S / W = 202ms

44. But TCP setup … One round trip for the server

    and client to say hello.

45. TCP Setup T = time S = payload size W

    = bandwidth (bytes/s) P = one-way transmission time (ping) RTT = round-trip time (2 * P) S = 78K W = 1500K/s P = 75ms RTT = 2 * P = 150ms T = (2 * RTT) + S / W = 352ms

46. But TCP slow start … A server starts by sending

    only a few packets of data, increasing the “congestion window” as packets are acknowledged.

47. TCP Setup T = time S = payload size W

    = bandwidth (bytes/s) P = one-way transmission time (ping) RTT = round-trip time (2 * P) C = congestion window (bytes) R = number of round trips S = 78K W = 1500K/s P = 75ms C = 14.6K RTT = 2 * P = 150ms R = ceil( log_2( (S / C) + 1) ) = 3 T = ( RTT * (1 + R) ) + S / W = 652ms - excludes DNS lookup for initial request to a host - assumes instantaneous server response, no packet loss, 
 and lots of other things

48. Easy Math T = time S = payload size W

    = bandwidth (bytes/s) S = 78K W = 1500K/s T = S / W = 52ms

49. Reality T = time S = payload size W =

    bandwidth (bytes/s) P = one-way transmission time (ping) RTT = round-trip time (2 * P) C = congestion window (bytes) R = number of round trips S = 78K W = 1500K/s P = 75ms C = 15K RTT = 2 * P = 150ms R = ceil( log_2( (S / C) + 1) ) = 3 T = ( RTT * (1 + R) ) + S / W = 652ms - excludes DNS lookup for initial request to a host - assumes instantaneous server response
50. None
51. None

52. 14.6K 43.8K 102.2K

53. 1K T = time S = payload size W =

    bandwidth (bytes/s) P = one-way transmission time (ping) RTT = round-trip time (2 * P) C = congestion window (bytes) R = number of round trips 1K file size 75ms ping 15K congestion window 1500K/s bandwidth RTT = 150 R = 1 T = ( 150 * (1 + R) ) + 1 = 301ms

54. 43.8K Maximum file size to ensure only 2 round trips

    on a new connection.

55. 14.6K Maximum file size to ensure only 1 round trip

    
 on a new connection.

56. which workflows matter most? where are the bytes coming from?

    how can we hold the line after improving?
57. None

58. Framework or no, 
 JS growth & complexity 
 is

    unbounded 
 without constraints.

59. Education is 
 a powerful enabler 
 of improvement.

60. It’s unrealistic for you […] 
 to learn a framework

    
 that solves a problem 
 you’ve never experienced. 
 Tessa Thornton,
 How to Learn Web Frameworks https://medium.com/shopify-ux/how-to-learn-web-frameworks-9d447cb71e68

61. Every technology decision is eventually regrettable.

62. Building 
 for the web
 is better today.

63. I help
 people
 get jobs. @rmurphey rmurphey.com