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

Site-Speed That Sticks

Avatar for Harry Roberts Harry Roberts
November 14, 2024
Technology
11
880

Site-Speed That Sticks

Avatar for Harry Roberts

Harry Roberts

November 14, 2024
Tweet

More Decks by Harry Roberts

See All by Harry Roberts
How to Think Like a Performance Engineer
Avatar for Harry Roberts csswizardry
27
2k
cache rules everything
Avatar for Harry Roberts csswizardry
5
3.5k
My Website Is Slow!
 Where Do I Start?
Avatar for Harry Roberts csswizardry
5
510
Optimising Largest Contentful Paint
Avatar for Harry Roberts csswizardry
37
3.4k
Get Your Head Straight
Avatar for Harry Roberts csswizardry
15
20k
From Milliseconds
 to Millions: A Look at the Numbers Powering Web Performance
Avatar for Harry Roberts csswizardry
1
2.6k
More Than You Ever Wanted to Know About Resource Hints
Avatar for Harry Roberts csswizardry
6
9.4k
It’s My (Third) Party, and I’ll Cry if I Want To
Avatar for Harry Roberts csswizardry
13
5.6k
FaCSSt: CSS & Performance
Avatar for Harry Roberts csswizardry
26
4.1k

Other Decks in Technology

See All in Technology
20201008_ファインディ_品質意識を育てる役目は人かAIか___2_.pdf
Avatar for ファインディ株式会社（イベント資料アップ用） findy_eventslides
1
510
[2025-09-30] Databricks Genie を利用した分析基盤とデータモデリングの IVRy の現在地
Avatar for 和田 悠佑 wxyzzz
0
500
about #74462 go/token#FileSet
Avatar for tom twinkle tomtwinkle
1
420
Goに育てられ開発者向けセキュリティ事業を立ち上げた僕が今向き合う、AI × セキュリティの最前線 / Go Conference 2025
Avatar for GMO Flatt Security flatt_security
0
350
データエンジニアがこの先生きのこるには...?
Avatar for 10xinc 10xinc
0
450
Trust as Infrastructure
Avatar for Bryan Cantrill bcantrill
0
350
BirdCLEF+2025 Noir 5位解法紹介
Avatar for MYSO myso
0
200
Findy Team+のSOC2取得までの道のり
Avatar for adachi.ryo rvirus0817
0
370
AIが書いたコードをAIが検証する！自律的なモバイルアプリ開発の実現
Avatar for henteko henteko
1
350
PLaMoの事後学習を支える技術 / PFN LLMセミナー
Avatar for Preferred Networks pfn
PRO
9
3.9k
AWSにおけるTrend Vision Oneの効果について
Avatar for shima shimak
0
130
【新卒研修資料】LLM・生成AI研修 / Large Language Model・Generative AI
Avatar for BrainPad brainpadpr
25
17k

Featured

See All Featured
ReactJS: Keep Simple. Everything can be a component!
Avatar for Pedro Nauck pedronauck
667
120k
Six Lessons from altMBA
Avatar for Skipper Chong Warson skipperchong
28
4k
The Myth of the Modular Monolith - Day 2 Keynote - Rails World 2024
Avatar for Eileen M. Uchitelle eileencodes
26
3.1k
4 Signs Your Business is Dying
Avatar for Josh Pigford shpigford
185
22k
RailsConf 2023
Avatar for Aaron Patterson tenderlove
30
1.2k
Practical Tips for Bootstrapping Information Extraction Pipelines
Avatar for Matthew Honnibal honnibal
PRO
23
1.5k
The Pragmatic Product Professional
Avatar for Laura Van Doore lauravandoore
36
6.9k
Designing Experiences People Love
Avatar for Jonathan Moore moore
142
24k
The MySQL Ecosystem @ GitHub 2015
Avatar for Sam Lambert samlambert
251
13k
Typedesign – Prime Four
Avatar for Hannes Fritz hannesfritz
42
2.8k
What’s in a name? Adding method to the madness
Avatar for Product Marketing Alliance productmarketing
PRO
23
3.7k
GraphQLの誤解/rethinking-graphql
Avatar for sonatard sonatard
73
11k

Transcript

  1. site-speed that sticks

  2. None

  3. hi, i’m harry

  4. None

  5. five key topics

  6. 1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook

  7. 1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook

  8. 1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook

  9. 1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook

  10. 1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook

  11. 1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook

  12. metrics

  13. not all metrics are born equal

  14. different metrics for different people on different occasions with different

    levels of disclosure

  15. kpis, enablers, predictors

  16. kpis

  17. definition + target

  18. what are we working toward?

  19. of interest to the business

  20. core web vitals

  21. ‘which number on which dashboard of which service?’

  22. “We want a one-second improvement in Largest Contentful Paint.” —

    My Client
  23. None
  24. None
  25. None

  26. enablers

  27. metrics that directly influence kpis

  28. of interest to engineering teams

  29. ttfb, input delay

  30. predictors

  31. signals of good/bad performance

  32. highly quantitative

  33. of interest to engineers

  34. bundle size, long tasks, blocking css

  35. great for root-causing and reverse engineering

  36. localhost

  37. localhost is: seldom live-like, pretty dang fast, un-bundled

  38. know your tools inside out

  39. None
  40. None

  41. csswz.it/perfnow25

  42. one weird trick…

  43. None

  44. // plugins/delay.server.ts export default defineNuxtPlugin(async () => { await new

    Promise(resolve => setTimeout(resolve, 900)) })
  45. None
  46. None

  47. <head> <link rel=stylesheet href=https://slowfil.es/file?type=css&delay=800> </head>

  48. core web vitals are too big for localhost

  49. if you’re working locally, measure locally

  50. bare-metal metrics

  51. None
  52. None

  53. 1

  54. 1 2

  55. 1 2 3

  56. 1 2 3

  57. None
  58. None

  59. 1

  60. 1 2

  61. 1 2 3

  62. 1 2 3

  63. external: 1842ms inlined: 1250ms

  64. None

  65. these are very private metrics

  66. backstops

  67. …and budgets

  68. what is the worst possible performance we will accept?

  69. set it to the worst reading in the last release

    cycle
  70. None

  71. this is where synthetic testing comes into it

  72. synthetic testing; real user monitoring

  73. when to fail a release

  74. None

  75. predictors as tripwires

  76. None

  77. budgets versus targets

  78. budgets are backstops; targets are ambitions

  79. target == kpi

  80. None

  81. monitoring

  82. the m in rum stands for monitoring

  83. “Insanity is doing the same thing over and over again

    and expecting different results.” — Rita Mae Brown
  84. None

  85. 🎉

  86. ?

  87. None
  88. None
  89. None
  90. None

  91. it’s the exact same file

  92. None

  93. you’re monitoring variation in tests

  94. None

  95. only alert on your kpis

  96. None
  97. None

  98. 0.9952409649

  99. None
  100. None

  101. always follow the numbers

  102. playbook

  103. “Fighting regressions took priority over optimizations […]” — Michelle Vu,

    Pinterest
  104. None

  105. it’s all for nothing if you don’t have a plan

  106. response = f(severity, duration)

  107. severity

  108. acceptable: <10%

  109. moderate: 10–25%

  110. severe: 25–50%

  111. critical: >50%

  112. duration

  113. temporary: 24–48hr

  114. sustained: >48hr

  115. long-term: >1 release cycle

  116. unresolved: many release cycles

  117. a kpi regression of over 10% for one week requires

    remediation in the next sprint

  118. a kpi regression of over 100% for one hour requires

    rollback immediately

  119. a kpi regression of over 25% for one day requires

    remediation in the current sprint

  120. an enabler regression of over any% for any time needs

    the team’s attention over the next sprint

  121. a predictor regression of over any% for any time needs

    my attention over the next sprint

  122. you need a framework to fill in these blanks

  123. early triage

  124. who, what, when, where, and why?

  125. what?

  126. what has regressed?

  127. when?

  128. when did it start? is it still like that?

  129. where?

  130. is it a business-critical part of the site?

  131. who?

  132. who owns the problem?

  133. why?

  134. can you conduct early triage?

  135. None
  136. None
  137. None
  138. None
  139. None
  140. None

  141. key takeaways

  142. increase confidence

  143. use the right tool for the right job

  144. have a plan of attack

  145. agree; commit

  146. thank you

  147. harry.is/for-hire