Robin Marx

illustration of 3 cars driving down a 3 lane road, carrying web resources

Day 011 #100DaysOfPerf: I shared the hpbn book (Day 002) that wall about levelling up on HTTP. Well, let's look at CURRENT HTTP data from the @httparchive.org's Web Almanac. On deck ✨ HTTP ✨ by none other than @programmingart.bsky.social. Let's take a look: 🧵⬇️
almanac.httparchive.org/en/2024/http

Featured posts this week:
⚡ Fast and furious: @catchpointsystems.bsky.social
⚡ How browsers REALLY load Web pages: @programmingart.bsky.social
⚡ Client-Side Rendering: @adevnadia.bsky.social
⚡ More Edge features get a perf boost: Mohamed Mansour
Special thanks to the authors.🙏

GitHub - mdn/content: The content behind MDN Web Docs The content behind MDN Web Docs. Contribute to mdn/content development by creating an account on GitHub.

The "contributing" options on the web.dev bottom for example don't seem quite as inviting as say github.com/mdn/content, which imo helps reinforce the impression this is a google/chromium-first thing.

(just trying to provide some feedback on why I had my initial impressions on web.dev)

Arguably, you can't expect Chrome engineers to add non-chromium behaviours in there, but it's also unclear (at least to me on the outside looking in) if you would accept contributions from say Firefox/Safari/outsiders to extend that type of article with cross-browser content (it's already extensive)

Optimize resource loading with the Fetch Priority API  |  Articles  |  web.dev The Fetch Priority API indicates the relative priority of resources to the browser. It can enable optimal loading and improve Core Web Vitals.

Good to know this is the intent (as @paul.kinlan.me also echoed), but from my perspective, web.dev articles that talk about the type of topics I research, are usually still exceptionally chromium-focused, even for features in baseline (prime example being web.dev/articles/fet...).

web.dev Guidance to build modern web experiences that work in every browser.

Thanks Patrick. I would definitely be interested in contributing docs, I'm just not sure where they would fit best. Web.dev is traditionally google/chromium-focused and not sure if they'd want to change. MDN doesn't seem to do much of this more "fuzzy" content in my experience, but maybe I'm wrong?

Sadly yes... Dinner date with the wife I'm afraid

It was just bit.ly/loadline :)

We kick off with @programmingart.bsky.social showing us how browsers REALLY load web pages to a full room! 🤡 fosdem.org/2025/schedul...

HTTP/1.1/2/3 are all present in unison on the web today, w/ a 21/70/9 split. Or is it? @programmingart.bsky.social will shed the light on the @httparchive.org Web Almanac data, and the reality of the protocol's adoption. Join us to hear him share findings this Thursday
🔗 ⬇️
bsky.app/profile/henr...

Making Chrome QUICer In October 2020, Chrome enabled HTTP/3 by default . HTTP/3 ( RFC 9114 ) runs over IETF QUIC ( RFC9000 ). Default-enabling HTTP/3 in Chrome ...

Hmm, interesting - looks like Chrome added support for the ORIGIN frame for HTTP/3 (eliminating the need for a DNS lookup before coalescing connections).

blog.chromium.org/2024/12/maki...

Getting Real (small) With Compression Dictionaries Compression dictionary transport is a relatively new feature in HTTP that allows for using custom compression dictionaries to improve the compression of HTTP responses. The results can be pretty drama...

Web performance calendar day 29 with @patmeenan.com on how to ship a production-quality dictionary compression implementation to reduce HTML/JS/CSS response sizes dramatically (as in 60-90% dramatically!)

calendar.perfplanet.com/2024/getting...

Unattributed Navigation Overhead (UNO) I promise you, UNO (Unattributed Navigation Overhead) is the best Christmas present you will get this year! UNO, not the card game, is the missing piece for many of your unsolved performance puzzles...

Web performance calendar day 21 with @timvereecke.bsky.social giving us the gift of UNO (Unattributed Navigation Overhead) - an insight into the blind spot gaps between TTFB components

calendar.perfplanet.com/2024/uno/

Let's hope we can re-invigorate that then :) Thanks for sharing!

Top 8 things I want in the Devtools Network Panel Look, I'm just going to come out and say it: I'm a jealous man. I'm not proud of it, but I'm a jealous man. These past months and years, I've seen great update after amazing update (not you) being rel...

For this year's #webperf calendar, I wrote "Top 8️⃣ things I want in the Devtools Network Panel".

I often struggle to debug modern networking features in browsers (like DNS HTTPS records, 0-RTT, and Happy Eyeballs 👀) and have some ideas on how to improve that!

calendar.perfplanet.com/2024/top-8-t...

Hmz that's indeed much lower than I would have expected... Maybe @tunetheweb.com can have a look once he's less busy? 😇

But as a percentage of all sites that use preload (the ~2.5 million):

235,535 / 2,506,138 = 0.09 = 9%
228,895 / 2,506,138 = 0.09 = 9%
so about 18% combined (round up to say 20% for the leftovers that uselessly preload more than 2 resources).

Hope that helps :)

Hm, I think it's more like 20% :) Looking at the raw results (docs.google.com/spreadsheets...).

2,506,138 used preload (~20 of all desktop homepages)
235,535 has 1 unused preload (~1.9% of all desktop homepages)
228,895 has 2 unused preloads (~1.85% of all)

Combined <4% of total (as reported)

My meme-game is on-point for my (first) perfcalendar 2024 submission @stoyan.org!

MASSIVE thank you to @tunetheweb.com, without whom there would have been no data to write about in the first place!

Barry is consistently one of the kindest and most hardworking people in #webperf (and beyond) and deserves our gratitude!

HTTP | 2024 | The Web Almanac by HTTP Archive The HTTP chapter of the 2024 Web Almanac covers data on historical versions of HTTP used across the web, as well as the uptick in adoption of HTTP/2 and HTTP/3 from alt-svc and DNS HTTPS records. Addi...

This year, I again had the honour of authoring the HTTP chapter of the Web Almanac: almanac.httparchive.org/en/2024/http

It's full of interesting stats on technologies like HTTP/3, DNS HTTPS records, preloads, 103 Early Hints and the FetchPriority API.

It also has Pirates 🏴‍☠️ and Marry Poppins 🌂!

HTTP | 2024 | The Web Almanac by HTTP Archive The HTTP chapter of the 2024 Web Almanac covers data on historical versions of HTTP used across the web, as well as the uptick in adoption of HTTP/2 and HTTP/3 from alt-svc and DNS HTTPS records. Addi...

The HTTP chapter was written by @programmingart.bsky.social and being our first name, is a subject close to our heart 😀

almanac.httparchive.org/en/2024/http

Read it for lots of fascinating deep dive into to how web resources are delivered and the latest on how developers can influence that!

the blog post writeup of this is so good eieio.games/blog/writing...

Release v12.2.3 · GoogleChrome/lighthouse Full Changelog This release is only for PageSpeed Insights and we expect it to ship within 2 weeks. Notable Changes The CPU throttling factor for PageSpeed Insights has been adjusted (#16256) to ac...

Some of you complain that PageSpeed Insights is a bit unrealistically show for the Lighthouse test.

The PSI servers are a bit underpowered so we're changing the throttling, which should make the Performance score go up when this rolls out in the next few weeks.

github.com/GoogleChrome...

🧵 1/5

HTMHell Advent Calendar 2024 - HTMHell An article, talk, or tool that focuses on HTML every day until Christmas.

It's December 1st, and you know what that means: For the next 24 days, we'll publish articles about HTML by 24 authors. You can read them on the website or subscribe to the RSS channel.

Enjoy!

htmhell.dev/adventcalend...

YouTube video by We Love Speed Beyond the basics, pushing the limits of webperf — Tim Vereecke

My #wls2024 talk is available!

I delved into advanced tuning techniques to make a modern website lightning fast.

Using lots of charts and data, I showed how they helped make scalemates the fastest (and largest) modeling website in the world.

www.youtube.com/watch?v=rjUr...

#webperf

Not that I know of no... Would actually be quite interesting to have that as a signal imo... @tunetheweb.com @patmeenan.com

It's Fastly and QUIC.cloud, as detailed in the paper linked on the bottom of that slide: jherbots.info/public_media...

Note that this is only for HTTP/3. I also have some tests for HTTP/2 (which is still equally broken on some servers) but haven't made anything public for that yet.

1. Yeah, could be, didn't really dig super-deep into this one yet because it felt so inconsistent.

2. this was actually SO MUCH HARDER THAN IT LOOKS. I was practicing in my hotel room all night before... luckily I think @guaca.bsky.social has some skills here, so you can book us as a double act ;)

Here, I think Safari just falls back to normal tight mode logic: it's still in tight mode, but there's just 1 high-prio resource in flight (the 3rd CSS that's slow to load), so it starts loading low-prio stuff, but only 1 at a time.

I don't think this is a special case due to stylesheets.