Performance and accessibility aren’t features that can linger at the bottom of a Jira board to be considered later when it’s convenient.
Instead we must start to see inaccessible and slow websites for what they are: a form of cruelty. And if we want to build a web that is truly a World Wide Web, a place for all and everyone, a web that is accessible and fast for as many people as possible, and one that will outlive us all, then first we must make our websites something else altogether; we must make them kind.
Harry takes a deep dive into the performance metric of “time to first byte”, or TTFB if you using initialisms that take as long to say as the thing they’re abbreviating.
This makes a great companion piece to Drew’s article on server timing headers.
Fast software is not always good software, but slow software is rarely able to rise to greatness. Fast software gives the user a chance to “meld” with its toolset. That is, not break flow.
Scott re-examines the browser support for loading everything-but-the-critical-CSS asynchronously and finds that it might now be as straightforward as this one declaration:
<link rel="stylesheet" href="/path/to/my.css" media="print" onload="this.media='all'">
I love the fact the Filament Group are actively looking at how deprecate their
loadCSS polyfill—exactly the right attitude for polyfills in general.
Time to Interactive (TTI) is the most impactful metric to your performance score.
Therefore, to receive a high PageSpeed score, you will need a speedy TTI measurement.
At a high level, there are two significant factors that hugely influence TTI:
Trust no one! Harry enumerates the reason why you should be self-hosting your assets (and busts some myths along the way).
There really is very little reason to leave your static assets on anyone else’s infrastructure. The perceived benefits are often a myth, and even if they weren’t, the trade-offs simply aren’t worth it. Loading assets from multiple origins is demonstrably slower.
Page web bloat score (WebBS for short) is calculated as follows:
WebBS = TotalPageSize / PageImageSize
Yes, this is a tongue-in-cheek somewhat arbitrary measurement, but it’s well worth reading through the rationale for it.
How can the image of a page be smaller than the page itself?
This is a very useful new feature in Calibre, the performance monitoring tool. Now you can get data about just how much third-party scripts are affecting your site’s performance:
The best way of circumventing fear and anxiety around third party script performance is to capture metrics that clearly articulate their performance impact.
Following on from Harry’s slides, here’s another round-up of those
rel attribute values that begin with
Slides from Harry’s deep dive into
What happens when you’re AMP pages are slower than your regular pages …but you’re forced to use AMP anyway if you want to appear in the top stories carousel.
AMP isn’t about speed. It’s about control.
The elephant in the room here is pre-rendering: that’s why Google aren’t using page speed alone as a determining factor for what goes in the carousel.
I love everything about this article and I can’t wait for part two.
What we tend to forget is that the environment websites and web apps occupy is one and the same. Both are subject to the same environmental pressures that the large gradient of networks and devices impose. Those constraints don’t suddenly vanish when we decide to call what we build “apps”, nor do our users’ phones gain magical new powers when we do so.
Needless to say, I endorse this message:
53% of mobile visits leave a page that takes longer than 3 seconds to load. That means that a large number of visitors probably abandoned these sites because they were staring at a blank screen for 3 seconds, said “fuck it,” and left approximately half way before the page showed up. The fact that the next page interaction would have been quicker—assuming all the JS files even downloaded correctly in the first attempt—doesn’t amount to much if they didn’t stick around for the first page to load. What was gained by putting the business logic in the front end in this scenario?
I think I physically winced on more than one occassion as I read through Jake’s report here.
He makes an interesting observation at the end:
However, none of the teams used any of the big modern frameworks. They’re mostly Wordpress & Drupal, with a lot of jQuery. It makes me feel like I’ve been in a bubble in terms of the technologies that make up the bulk of the web.
Yes! This! Contrary to what you might think reading through the latest and greatest tips and tricks from the front-end community, the vast majority of sites out there on the web are not being built with React, Vue, webpack or any other “modern” tools.
My first reaction to this was nervousness. Of all the companies to trust with intercepting and rerouting page requests, Google aren’t exactly squeeky clean, what with that whole surveillance business model of theirs.
Still, this ultimately seems to be a move to improve the end user experience, and I’m glad to see this clarification:
Lite pages are only triggered for extremely slow sites, so we encourage developers to measure how well their pages are currently performing over slow networks.
Lite pages as a badge of shame (much like AMP in my eyes).
I like Tim’s definition here:
A performance budget is a clearly defined limit on one or more performance metrics that the team agrees not to exceed, and that is used to guide design and development.
And I agree about the four attributes required for a performance budget to succeed. It must be:
The point is not to let the performance budget try to stand on its own, somewhere hidden in company documentation collecting dust. You need to be proactive about making the budget become a part of your everyday work.
This is a really clever technique from Scott that he unveiled at An Event Apart in Seattle. It uses a header sent by a service worker to distinguish between returning and new visitors—much neater than relying on a cookie. I’ve updated my service worker on The Session to use this technique now.
Harry breaks down
cache-control headers into steps that even I can understand. I’ll be using this a reference for sure.