Web Performance 101 (opens in new tab)

> The best way to make your site ultra-responsive is to cut out all the bloat.

And if the site doesn't do it, you can often decrease page load time by turning on your browser's built-in tracking protection:

https://blog.mozilla.org/firefox/tracking-protection-always-...

It's a bit sad that blocking trackers can cut page load time in half but that is unfortunately the web we have.

This part is like always missing from such presentations. First point always should be remove as much as possible.

Exactly. The weird thing is that you don't have to loose that JS advantages if you pull out an old school technique called progressive enhancement. If you mix that with a strict no-js first design you will achieve performance that no ultra optimised "dynamic" style can compete with.

I have noticed a lot more landing pages using gradients/plain backgrounds instead of a big hero image.

"Removing useless cruft" doesn't seem very practical, did it often work for you? In my experience, erring on the side of making existing stuff work faster is better than removing non-critical features.

> and then you can still apply such techniques to whatever is left to squeeze out a bit more improvement

Yes, like "compressing/minifying/whatever else", which is the subject of this presentation.

What's next? Really, before minifying, you have to get a server to serve anything in the first place. And register a domain name. And, hmmm, well first you need to buy a computer and plug it in...

I found this super interesting and there were a few areas I wasn't familiar with where I learnt something new. Kudos.

Meta node: I find this presentation style to be great. You can scroll up and down at speed, can text search the whole page, we have a nice mixture of imagery and text, its clean and accessible, we even have a table of contents. How did you author this?

Some more ideas for your next talk:

HTTP2 is a doddle to implement. Do it.

PWA makes it possible to work fully offline.

With CSS best to chuck it all out including those reset files someone wrote a decade ago. Instead re-write the whole lot using CSS Grid and using custom variables.

Inline the SVG into the CSS as custom variables.

Get rid of the bulk of the JS by only targeting evergreen browsers. No polyfills, no jQueries, just minimal javascript that does a lot of things in the PWA and manipulates CSS custom variables rather than the DOM.

Use HTML5 properly, with no lip service. Get rid of JS for forms and rely on HTML5 to do it.

Pagespeed to sort out the images and make them into source sets.

The goal of a lot of the above is to strip out convoluted build tools and have actual neat HTML that can be maintained. No more 'add only' CSS to hand on to the next guy, instead have something with comments in the code and sensible names that target HTML5 things like 'main' and 'aside' or 'nav' rather than made up class names.

A final thought is that the starting point can be to build a green website, i.e. one that cause too much cruft to be downloaded. This is the same thing as 'minimizing/cutting out bloat' but I find that setting out to build a website that sets the example of being green is a better mindset than 'must do those hacky things to make website faster'.

Good article! The main thing I didn't quite agree with was the blanket statement that GIFs should never be used. I fully agree that they shouldn't be used for video clips, and I understand this was the main context, but for small animations (especially pixel-art types) they're still a good format.

I also found them useful recently for small (28x28px) thumbnail images on my personal website. On average, saved as a PNG the thumbnails were 20kb, as a JPEG 9kb, and as an optimized GIF about 1-2kb. With about 100 thumbnails on one of the pages, the savings are pretty significant. (At least, this seemed to be the best approach; if anyone with more knowledge of image compression has a better suggestion, please let me know).

I'm saddened to not see Closure listed on there for JS minification. It might be worth mentioning that there are more advanced minification tools.

Also, there doesn't seem to be anything on JSON minification, which is a sizable portion of responses. There are techniques to transpose JSON objects to be easier to gzip compress.

A few slides after you say that gifs should never be used, you use a gif ;)

I definitely learned something, thanks

> These tools are no good for automatic lossy image compression

The self-promotion didn't bother me until this claim, because you posted some great advice along with it.

ImageOptim is great. If you choose "lossy minification" it does automatic lossy image compression, preserving perceptual image quality while making huge reductions to file sizes. Users can even adjust how aggressive it is.

I'll take your word for it that I could get 13-24% smaller file sizes with your Optimage product on top of the 80% (or whatever) that I can get with ImageOptim. But I'd prefer that you didn't claim that other choices are "no good".

> [1] https://getoptimage.com/benchmark

A score of 24/55 for TinyPNG and then 55/55 for their own service makes it look as if this article is an advertisement. Especially since TinyPNG gets better/very similar file size while staying visually lossless up to a point (images where it's nothing but a bunch of rainbow gradients are its weakness).

Remember that TinyPNG is optimized for web use where artifacts are tolerated. It was configured with that in mind. They test for images that are visually identical and won't get it from any images optimizer that is made for web usage.

Users only spend a few seconds looking at images that on web pages and the artifacts from optimizers are very minor. See: https://3perf.com/talks/web-perf-101/#images-compress-jpg-si...

Why? Most trackers load async and don’t block the page from loading or rendering. I get it’s popular to hate but technically they load without blocking page from rendering. A good read is https://sites.google.com/a/webpagetest.org/docs/using-webpag...

It slows things down especially because the browser will likely have to layout/re-render the page everytime a new chunk of CSS arrives.

Generally avoid splitting CSS. Even if you don't use all your CSS on every page, a cached 100kb CSS file will outperform a bunch of unique-per-page 25kb CSS files everytime (especially since it's 0 requests for the second page). Except for dial-up probably.

It may be beneficial to split CSS files somewhere above the 300kb mark, but I wouldn't know. My one-page-app is only about 500kb over the wire, including CSS, JS, Fonts and HTML. ~30KB of that is CSS.

I've been optimizing that for years though.

Http/2 makes this less important. Perhaps still worth it for CSS, but higher effort things like image spriting might not be as attractive now.

Back in the day, reducing total HTTP requests meant less time waiting through the network lag (round trip times). With HTTP2, that's less important because it allows multiple requests simultaneously on the same connection (fewer network connections made=less lag). It's a nice to have feature and has measurable benefits, but I don't rely on it. There is a non-negligible cost for the browser to parse yet another CSS/JS file, but I'm not sure exactly what those costs are.

This is the correct way to do a lot of it. For instance, images. Yes you could optimise the things in Photoshop but actually you want your artworkers to be doing the images right, so they look good and tell the story. Optimising the images is something that should be abstracted out, so on your dev version of the site everything is in 100% glorious, maybe even phone-res megapixels.

Then, abstracted out with PageSpeed you can deliver the webp when you need to and also the source set images so every device has the right size images that update themselves automagically if people zoom in.

Same with minification, why have complex build tools when you can just have PageSpeed do it properly?

You can also have beautiful HTML for view source by putting on the right PageSpeed filters.

The list goes on, apart from the results it also does the abstraction bit, so artworkers can do their Photoshop stuff unencumbered, same for frontend and backend devs.

The thing about it though is that you need to understand a mix of different things that are nowadays split up into different job roles of ever increasing specialisation. A Photoshop person isn't going to go all command line on the server for the perfect PageSpeed Nginx setup, neither is a CSS person, a UX expert, a JS expert or a backend expert. Not even the guy who keeps the site online is going to typically step up to using Pagespeed for the benefit of the team. Pagespeed just doesn't fit into one of these niche-jobs so it is more likely to be found on smaller one-person efforts where there aren't the organisational hurdles in the way.

I've seen that. My understanding is that you can compress CSS, images, pretty much anything that does not require a secret. So if your sending a secret to get an image, I think you're doing something wrong. If you keep the secret in a cookie, or transfer it in a header rather than a body, you should be clear to use Gzip and HTTP compression.