Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Site-Speed That Sticks
Search
Harry Roberts
November 14, 2024
Technology
10
660
Site-Speed That Sticks
Harry Roberts
November 14, 2024
Tweet
Share
More Decks by Harry Roberts
See All by Harry Roberts
How to Think Like a Performance Engineer
csswizardry
24
1.7k
cache rules everything
csswizardry
5
3.4k
My Website Is Slow! Where Do I Start?
csswizardry
5
480
Optimising Largest Contentful Paint
csswizardry
37
3.3k
Get Your Head Straight
csswizardry
15
20k
From Milliseconds to Millions: A Look at the Numbers Powering Web Performance
csswizardry
1
2.5k
More Than You Ever Wanted to Know About Resource Hints
csswizardry
6
9.3k
It’s My (Third) Party, and I’ll Cry if I Want To
csswizardry
13
5.5k
FaCSSt: CSS & Performance
csswizardry
26
4.1k
Other Decks in Technology
See All in Technology
Liquid Glass革新とSwiftUI/UIKit進化
fumiyasac0921
0
240
Yamla: Rustでつくるリアルタイム性を追求した機械学習基盤 / Yamla: A Rust-Based Machine Learning Platform Pursuing Real-Time Capabilities
lycorptech_jp
PRO
3
140
PHPでWebブラウザのレンダリングエンジンを実装する
dip_tech
PRO
0
210
生成AIで小説を書くためにプロンプトの制約や原則について学ぶ / prompt-engineering-for-ai-fiction
nwiizo
4
2.5k
Tech-Verse 2025 Global CTO Session
lycorptech_jp
PRO
0
300
Understanding_Thread_Tuning_for_Inference_Servers_of_Deep_Models.pdf
lycorptech_jp
PRO
0
140
How Community Opened Global Doors
hiroramos4
PRO
1
120
なぜ私はいま、ここにいるのか? #もがく中堅デザイナー #プロダクトデザイナー
bengo4com
0
480
生成AI時代 文字コードを学ぶ意義を見出せるか?
hrsued
1
600
Fabric + Databricks 2025.6 の最新情報ピックアップ
ryomaru0825
1
140
セキュリティの民主化は何故必要なのか_AWS WAF 運用の 10 の苦悩から学ぶ
yoh
1
190
A2Aのクライアントを自作する
rynsuke
1
210
Featured
See All Featured
KATA
mclloyd
30
14k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
29
1.8k
Code Review Best Practice
trishagee
68
18k
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
31
1.2k
How to train your dragon (web standard)
notwaldorf
94
6.1k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
281
13k
The Language of Interfaces
destraynor
158
25k
Design and Strategy: How to Deal with People Who Don’t "Get" Design
morganepeng
130
19k
We Have a Design System, Now What?
morganepeng
53
7.7k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
45
7.5k
Statistics for Hackers
jakevdp
799
220k
Rails Girls Zürich Keynote
gr2m
94
14k
Transcript
site-speed that sticks
None
hi, i’m harry
None
five key topics
1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook
1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook
1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook
1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook
1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook
1. metrics 2. localhost 3. backstops 4. monitoring 5. playbook
metrics
not all metrics are born equal
different metrics for different people on different occasions with different
levels of disclosure
kpis, enablers, predictors
kpis
definition + target
what are we working toward?
of interest to the business
core web vitals
‘which number on which dashboard of which service?’
“We want a one-second improvement in Largest Contentful Paint.” —
My Client
None
None
None
enablers
metrics that directly influence kpis
of interest to engineering teams
ttfb, input delay
predictors
signals of good/bad performance
highly quantitative
of interest to engineers
bundle size, long tasks, blocking css
great for root-causing and reverse engineering
localhost
localhost is: seldom live-like, pretty dang fast, un-bundled
know your tools inside out
None
None
csswz.it/perfnow25
one weird trick…
None
// plugins/delay.server.ts export default defineNuxtPlugin(async () => { await new
Promise(resolve => setTimeout(resolve, 900)) })
None
None
<head> <link rel=stylesheet href=https://slowfil.es/file?type=css&delay=800> </head>
core web vitals are too big for localhost
if you’re working locally, measure locally
bare-metal metrics
None
None
1
1 2
1 2 3
1 2 3
None
None
1
1 2
1 2 3
1 2 3
external: 1842ms inlined: 1250ms
None
these are very private metrics
backstops
…and budgets
what is the worst possible performance we will accept?
set it to the worst reading in the last release
cycle
None
this is where synthetic testing comes into it
synthetic testing; real user monitoring
when to fail a release
None
predictors as tripwires
None
budgets versus targets
budgets are backstops; targets are ambitions
target == kpi
None
monitoring
the m in rum stands for monitoring
“Insanity is doing the same thing over and over again
and expecting different results.” — Rita Mae Brown
None
🎉
?
None
None
None
None
it’s the exact same file
None
you’re monitoring variation in tests
None
only alert on your kpis
None
None
0.9952409649
None
None
always follow the numbers
playbook
“Fighting regressions took priority over optimizations […]” — Michelle Vu,
Pinterest
None
it’s all for nothing if you don’t have a plan
response = f(severity, duration)
severity
acceptable: <10%
moderate: 10–25%
severe: 25–50%
critical: >50%
duration
temporary: 24–48hr
sustained: >48hr
long-term: >1 release cycle
unresolved: many release cycles
a kpi regression of over 10% for one week requires
remediation in the next sprint
a kpi regression of over 100% for one hour requires
rollback immediately
a kpi regression of over 25% for one day requires
remediation in the current sprint
an enabler regression of over any% for any time needs
the team’s attention over the next sprint
a predictor regression of over any% for any time needs
my attention over the next sprint
you need a framework to fill in these blanks
early triage
who, what, when, where, and why?
what?
what has regressed?
when?
when did it start? is it still like that?
where?
is it a business-critical part of the site?
who?
who owns the problem?
why?
can you conduct early triage?
None
None
None
None
None
None
key takeaways
increase confidence
use the right tool for the right job
have a plan of attack
agree; commit
thank you
harry.is/for-hire