Don't Forget The Network: Your App Is Slower Than You Think (RailsConf 2016)

When you look at your response times, satisfied that they are "fast enough", you're forgetting an important thing: your users are on the other side of a network connection, and their browser has to process and render the data that you sent so quickly. This talk examines some often overlooked parts of web applications that can destroy your user experience even when your response times seem fantastic. We'll talk about networks, routing, client and server-side VMs, and how to measure and mitigate their issues.

Transcript

1. don't forget the network your app is slower than you think
2. André Arko @indirect
3. The Ruby Way therubyway.io
7. Routing
8. Routing your app has this
9. Routing how does it work?
10. Development App You
11. Production People Router Server App App Router Server App App Router
12. Routing how slow is it?
13. Routing does it back up?
14. Request time
15. Request time not the time you measure
16. Request time wall-clock time from real clients
17. Request time make requests from around the world
18. Servers
19. Servers you have no idea what is going on
20. really.
21. Runtime lag
22. Runtime lag how do you tell you lost consciousness?
23. Runtime lag you have it.
24. Runtime lag you have it. how bad is it?
25. VM lag
26. VM lag do you have it?
27. VM lag do you check for it?
28. VM lag do you know how to check for it?
29. Metrics
30. Metrics are important
31. Metrics tell you what is happening
32. you rn →
33. Metrics convince you you understand
34. you later →
35. Averages convince you you understand
36. Averages are lie-candy for your brain
37. “Normal” 5 -5 -4 -3 -2 -1 0 1 2 3 4 0 0.1 0.2 0.3 0.4
38. “Normal” 5 -5 -4 -3 -2 -1 0 1 2 3 4 0 0.1 0.2 0.3 0.4
39. Real Life 5 -5 -4 -3 -2 -1 0 1 2 3 4 0 0.1 0.2 0.3 0.4
40. brendangregg.com
41. brendangregg.com
42. just heard “w e have a great average” →
43. The problem with averages: If you put one hand in a bucket of ice and the other in a bucket of hot coals, on average, you’re comfortable. Erik Michaels-Ober @sferik
44. Averages mask problems
45. 10 0 1 2 3 4 5 6 7 8 9 250 0 50 100 150 200
46. Graph the median
47. 10 0 1 2 3 4 5 6 7 8 9 250 0 50 100 150 200
48. Graph 95th percentile
49. 10 0 1 2 3 4 5 6 7 8 9 250 0 50 100 150 200
50. Graph 99th percentile
51. 10 0 1 2 3 4 5 6 7 8 9 1000 0 250 500 750
52. Aggregate graphs another average
54. Breakout graphs show each source
56. Seriously, do it Visualize your data
57. graphic by Schutz and Avenue, CC-Attribution-ShareAlike, taken from from the Wikipedia article on Anscombe's quartet
58. Average of X: 9 Average of X: 9 Average of X: 9 Average of X: 9
59. Average of Y: 7.50 Average of Y: 7.50 Average of Y: 7.50 Average of Y: 7.50
60. Average of X Average of Y Variance of X Variance of Y Correlation of X and Y Linear regression All four data sets have the same
61. Aggregate alerts more dead servers than alive servers
62. site’s up if any servers are up!
63. Breakout alerts first dead server not all the servers
64. So, in the end
65. don’t forget the network
66. user experience is all that matters after deploying
67. @indirect andre@arko.net Questions?