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

My Website Is Slow!
 Where Do I Start?

My Website Is Slow!
 Where Do I Start?

Working as a performance consultant, I tend to collaborate with many different clients over the course of a year. Two things they all have in common: they all know their website is slow, but they don’t know why or where.

When some sites can run to dozens of templates across thousands of pages, finding problem areas can feel like shooting in the dark, and making a start can feel completely paralysing. In this talk, we’ll look at some practical approaches, tools, and workflows to quickly and effectively determine where our primary liabilities lie, what they are, and—most importantly—which steps to take first.

Harry Roberts

May 24, 2023

More Decks by Harry Roberts

Other Decks in Technology


  1. Infoshare, Gdańsk – May 2023
    My Website Is Slow!

    Where Do I Start?
    Harry Roberts – @csswizardry

    View Slide

  2. Hi There!
    I’m Harry.
    Consultant Web Performance Engineer

    Leeds, UK



    View Slide

  3. View Slide

  4. “How do we know site speed is suffering?

    Customer complaints? Competitor benchmarking?
    Internal monitoring?”

    View Slide

  5. Where do I start?!

    View Slide

  6. Guess

    View Slide

  7. Guess
    Please don’t do this.
    This is a terrible idea. I would not recommend that at all

    I actually used to do this.

    Click around the site until I happened upon an issue.

    Day one was paralysing.

    Sure, I could make educated guesses.

    But any guess is not ideal.

    View Slide

  8. Client-Supplied URLs

    View Slide

  9. “Can you provide a list of key pages
    or flows for me to look at?”

    View Slide

  10. “Are there any parts of the site
    to avoid completely?”

    View Slide

  11. Client-Supplied URLs
    Let them do the work…
    They know their website better than I do—they tell me!

    Let’s think a level up from pages.

    Representative page types/templates.

    There are still going to be some blind spots.

    But this is the easiest place to start.

    View Slide

  12. View Slide

  13. Home: https://www.apple.com/

    Product: https://www.apple.com/iphone/

    Model: https://www.apple.com/iphone-14-pro/

    PDP: https://www.apple.com/shop/buy-iphone/iphone-14-pro

    SRP: https://www.apple.com/us/search/airtag

    View Slide

  14. View Slide

  15. View Slide

  16. View Slide

  17. View Slide

  18. Site-Speed Topography

    View Slide

  19. View Slide

  20. Glossary
    JFC, WTF?!
    ATF: Above the Fold

    CLS: Cumulative Layout Shift

    CrUX: Chrome User eXperience Report

    CWV: Core Web Vital

    FCP: First Contentful Paint

    FP/SR: First Paint/Start Render

    GA: Google Analytics

    LCP: Largest Contentful Paint

    P: th Percentile

    PDP: Product Details Page

    S(E)RP: Search (Engine) Results Page

    SI: Speed Index

    TBT: Total Blocking Time

    TTFB: Time to First Byte

    View Slide

  21. “The physical appearance of

    the natural features of an area of land,

    especially the shape of its surface.”

    View Slide

  22. View Slide

  23. Clusters of Red or Green
    Overall page health.
    Horizontal runs of red show the page performs poorly in many areas.

    Green shows the opposite.

    Red at the start (TTFB) usually means red later on (e.g. LCP).

    View Slide

  24. View Slide

  25. Green to Red
    Good to bad…
    Vertically, how do individual metrics perform?

    Patterns usually emerge, but does anything buck the trend?

    View Slide

  26. View Slide

  27. Variability
    Our most troublesome metrics
    TTFB is our most stable metric.

    LCP is our most volatile.

    Highlights potential underlying differences between page types.

    The same metric being highly variable usually has significant causes.

    View Slide

  28. View Slide

  29. Implied Issues
    Delta between TTFB and FCP is render-blocking resources (i.e. your ).

    Delta between FCP and LCP is inefficient LCP candidates (e.g. lazy-loaded).

    Positive delta between SI and LCP is late-loaded ATF content.

    Positive delta between FP and FCP may be web-font problems.

    Highly prone to false-positives—double check everything.

    View Slide

  30. Questions
    What does it imply?
    Home and Product pages generally healthy; PDP and SRP are generally poor.

    Home page has great FCP but the worst LCP—lazy-loaded image?

    SRP has the worst TTFB—expensive/inefficient queries?

    Why does the model page have the best LCP but the worst SI?

    Why are the PDP and SRP so slow to start render?

    SRP has the worst CLS—is it client rendered?

    View Slide

  31. View Slide

  32. Why is the homepage LCP so slow?

    View Slide

  33. View Slide

  34. Why is SRP first-byte so slow?

    View Slide

  35. View Slide

  36. Why does the Model page have poor SI?

    View Slide

  37. View Slide

  38. Largest != most important.

    View Slide

  39. speakerdeck.com/csswizardry/lcp

    View Slide

  40. Why does the PDP shift so much?

    View Slide

  41. View Slide

  42. Why does the SRP shift so much?

    View Slide

  43. View Slide

  44. Why are the PDP and SRP so much slower?

    View Slide

  45. View Slide

  46. https://www.apple.com/





    View Slide

  47. View Slide

  48. View Slide

  49. Problem: That was all based on assumptions.

    View Slide

  50. URLs


    Device type

    Network conditions

    View Slide

  51. URL Problem? Solution

    View Slide

  52. — unsplash.com/photos/nEJmnfCCPmI

    View Slide

  53. Google Analytics

    View Slide

  54. View Slide

  55. View Slide

  56. Behavior » Site Speed » Overview

    View Slide

  57. View Slide

  58. Problem: Load times from every page
    in every country in every browser on
    every device type, averaged.

    View Slide

  59. View Slide

  60. View Slide

  61. Behavior » Site Speed » Page Timings

    View Slide

  62. View Slide

  63. View Slide

  64. View Slide

  65. Problem: Load times from every page
    in every country in every browser on
    every device type, averaged.

    View Slide

  66. Site Speed is not a number.

    View Slide

  67. Site Speed is not a number.

    View Slide

  68. … » Site Speed » Page Timings » Distribution

    View Slide

  69. 30.26 + 46.40 = 76.66%

    View Slide

  70. 94.8%

    View Slide

  71. View Slide

  72. View Slide

  73. Problem: This will be gone on 1 July 2023

    View Slide

  74. View Slide

  75. View Slide

  76. Chrome User eXperience Report

    View Slide

  77. developer.chrome.com/docs/crux

    View Slide

  78. — developer.chrome.com/docs/crux/about/
    “The Chrome User Experience Report [CrUX] is a
    dataset that reflects how real-world Chrome users
    experience popular destinations on the web.”

    View Slide

  79. Problem: Only captures Chrome traffic;
    only captures Core Web Vitals.

    View Slide

  80. pagespeed.web.dev

    View Slide

  81. View Slide

  82. CrUX API

    View Slide

  83. $ curl -s --request POST 'https://


    key=' \

    --header 'Accept: application/json' \

    --header 'Content-Type: application/json' \

    --data '{"formFactor":"PHONE","url":"https://


    ["largest_contentful_paint", "first_input_delay",


    View Slide

  84. "largest_contentful_paint": {

    "histogram": [

    { "start": 0,

    "end": 2500,

    "density": 0.37880096751900411 },

    { "start": 2500,

    "end": 4000,

    "density": 0.31850380096751846 },

    { "start": 4000,

    "density": 0.30269523151347738 }


    "percentiles": { "p75": 4374 }



    View Slide

  85. Problem: You still need to decide the URLs.

    View Slide

  86. Problem: You need to graph the data.

    View Slide

  87. Treo

    View Slide

  88. #notAnAd

    View Slide

  89. View Slide

  90. View Slide

  91. View Slide

  92. View Slide

  93. View Slide

  94. View Slide

  95. View Slide

  96. Not one of those URLs was in our initial list!

    View Slide

  97. View Slide

  98. View Slide

  99. Back to WebPageTest!

    View Slide

  100. Search Console

    View Slide

  101. search.google.com

    View Slide

  102. Insights on a plate.

    View Slide

  103. Search Console » Core web vitals

    View Slide

  104. View Slide

  105. View Slide

  106. View Slide

  107. URL Groups

    View Slide

  108. www.site.com

    View Slide

  109. www.site.com/forum/

    View Slide

  110. www.site.com/forum/

    View Slide

  111. View Slide

  112. View Slide

  113. View Slide

  114. URL Groups are a bit of a mess…

    View Slide

  115. View Slide

  116. Back to WebPageTest!

    View Slide

  117. Summary

    View Slide

  118. URL Problem? Solution

    View Slide

  119. Gather data from various sources.

    View Slide

  120. Use smart shortcuts.

    View Slide

  121. Use new insights to inform future testing.

    View Slide

  122. Go looking for trouble.

    View Slide

  123. Problem URL Solution

    View Slide

  124. speakerdeck.com/csswizardry
    Thank You

    View Slide