Web Performance 101

315 points by iamakulov 7 years ago | 53 comments

IMHO instead of going for compressing/minifying/whatever else, it is far better to just remove the useless cruft in the first place --- and then you can still apply such techniques to whatever is left to squeeze out a bit more improvement. The best way to make your site ultra-responsive is to cut out all the bloat.

clouddrover 7 years ago | |

> 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.

C1sc0cat 7 years ago | | |

This is mostly about tuning for performance for SEO - its amazing in 2018 that sites are delivered with unoptimized images - its not rocket science to use Photoshop or run all the images through a cli based tool.

Matrixik 7 years ago | |

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

diafygi 7 years ago | | |

You're totally right, but unfortunately, I think that's because the person writing these types of presentations don't have control over the content or design of the website, so removing stuff is often impossible or way more work that these tips.

The core holdback for slow websites is usually political, not technical or lack of skill. These presentations usually just try to make the best of those team dynamics issues.

nicbou 7 years ago | | |

It reminds me of "reduce, reuse, recycle, in that order". It's better to reduce resource consumption than to reduce its impact.

LolNoGenerics 7 years ago | |

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.

blinky1456 7 years ago | |

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

nicbou 7 years ago | | |

Truth be told, I love big hero images amd especially videos. However, they're the largest performance problem on my otherwise text-based website, and they're purely aesthetic.

silvenon 7 years ago | |

"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.

magicalist 7 years ago | |

> 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...

iamakulov 7 years ago |

So, in summer, I gave an introductory talk into web performance. This is its textual version :)

The talk includes 94 slides with text about:

  — why web performance matters
  — how to optimize:
    — JS (async/defer, code splitting)
    — CSS (the critical CSS approach & tools for it)
    — HTTP/connection stuff (Gzip/Brotli, preloading, CDN)
    — Images (compressing images, webp, progressive vs baseline images)
    — Fonts (the `font-display` trick)
  — and what tools help to understand your app’s performance

Would love to hear your feedback :)

vladdanilov 7 years ago |

> Compress your JPG images with the compression level of 70‑80.

In practice some images can get noticeable artifacts even at around 90. Most of JPEG compressors always apply chroma subsampling which is often destructive on its own [1]. On the contrary, many hidpi images can be compressed at around 50.

> Use Progressive JPEG… Thanks to this, a visitor can roughly see what’s in the image way earlier.

That's not the point of using progressive JPEGs nowadays. The 10-200% decompression slowdown is for the 5-15% size reduction.

> Use Interlaced PNG.

Don't. Interlaced PNGs can easily be 1/3 bigger. There are better ways to show loading images, and it's already used on the website.

> webpack has image-webpack-loader which runs on every build and does pretty much every optimization from above. Its default settings are OK

> For you need to optimize an image once and forever, there’re apps like ImageOptim and sites like TinyPNG.

These tools are no good for automatic lossy image compression [1]. The default is mostly JPEG 4:2:0 at quality 75, PNG quantized with pngquant at settings as low as conscience allows, missing out many PNG reductions and optimal deflate, no separation between lossy and lossless WebP if at all, etc.

As a result, the images on the website have about 13-24% more to optimize losslessly.

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

runako 7 years ago |

Based on my first scan, I've bookmarked this for the next time I have a perf issue on a site.

One thing I didn't notice is that one of the biggest speedups is removing junk from the pages. That could be too many JS trackers, user-hostile videos, or whatever. IMHO it's an underrated skill for Web developers to be able to push back with cogent arguments when asked to ruin the performance of the sites they work on.

taf2 7 years ago | |

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...

pmichalina 7 years ago | | |

Even just one poorly designed library can cause serious memory issues and trigger a ton of events. Usually it’s the marketing and ad people that throw around this “but they are async loaded”. That’s true, but doesn’t change the the fact that 40 trackers and their dependencies that come with it are slowing down and infringing on the users privacy. Let’s be real.

geekybiz 7 years ago | | |

They may load async but JS parse + compile + execute takes up client device CPU cycles and can block the main thread from responding back to user actions immediately. More so on mobile devices. This, and the fact that a typical page has multiple such trackers can cause notable overhead.

jazzyjackson 7 years ago | | |

They may not be blocking, but it's an extra burden on my limited bandwidth (slows other connections down while it downloads) and balloons the memory usage of sites that are otherwise delivering static content.

firasd 7 years ago |

Good document! I wanted to mention, something I've found to be significant in practical testing but often not considered by web developers is reducing the total number of HTTP queries. It seems like fetching 5 CSS files of 5kb each will significantly slow things down compared to combining them into one 25kb CSS file.

chmod775 7 years ago | |

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.

tyingq 7 years ago | |

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.

chrisweekly 7 years ago | | |

Yes -- in fact some webperf techniques are HTTP/1.1 - specific, and are actually _anti-patterns_ in HTTP/2. Spriting is one such example.

andreareina 7 years ago | | |

AFAIK SVG sprites (or inline SVG) are still required to style SVG images with external CSS.

theandrewbailey 7 years ago | |

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.

Klathmon 7 years ago |

I know i'm kinda self-promoting here, but since you mention webpack loaders like responsive-loader, and you recommend image-webpack-loader multiple times, I figured I could mention my plugin imagemin-webpack-plugin [0]

The problem with image-webpack-loader is that it only works on images which are `require`d or `import`ed. responsive-loader adds those images to webpack in a way that the loader cannot compress them.

Plus there are a bunch of other fany features that many helpful users have added like caching (no need to re-compress every image every time you run it!), the ability to minify images not in the webpack pipeline, and more.

[0] https://github.com/Klathmon/imagemin-webpack-plugin

Mr_Manager 7 years ago |

I've found the server PageSpeed module (in my case for Nginx) does a lot of good things "on the fly".

Theodores 7 years ago | |

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.

C1sc0cat 7 years ago | | |

No you need your artworkers to do it right it the flipping first place and not place excessive load on your live environment - or make the live environment more complex than it needs to be.

gitgud 7 years ago | | |

Well, in reality, all of these build tools for performance are design to deploy to a "static server" which is out of the control of the developer. That is why they process the images at build time rather than "on the fly"

Not saying that Page speed is wrong, but it is a niche tool that depends on the server side implementation. Some developers prefer to abstract "the server" from their architecture...

commandlinefan 7 years ago |

Careful with GZip and SSL, though - your payloads become vulnerable to the BREACH attack.

virmundi 7 years ago | |

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.

timmytwotime 7 years ago |

I would have liked to have seen more on HTTP/2