Conservative web development September 4, 2018 on Drew DeVault's blog

Today I turned off my ad blocker, enabled JavaScript, opened my network monitor, and clicked the first link on Hacker News - a New York Times article. It started by downloading a megabyte of data as it rendered the page over the course of eight full seconds. The page opens with an advertisement 281 pixels tall, placed before even the title of the article. As I scrolled down, more and more requests were made, downloading a total of 2.8 MB of data with 748 HTTP requests. An article was weaved between a grand total of 1419 vertical pixels of ad space, greater than the vertical resolution of my display. Another 153-pixel ad is shown at the bottom, after the article. Four of the ads were identical.

I was reminded to subscribe three times, for $1/week (after one year this would become $3.75/week). One of these reminders attached itself to the bottom of my screen and followed along as a scrolled. If I scrolled up, it replaced this with a larger banner, which showed me three other articles and an ad. I was asked for my email address once, though I would have had to fill out a captcha to submit it. I took out my phone and repeated the experiment. It took 15 seconds to load, and I estimate the ads took up a vertical space equal to 4 times my phone’s vertical resolution, each ad alone taking up half of my screen.

The text of the article is a total of 9037 bytes, including the title, author, and date. I downloaded the images relevant to the article, including the 1477x10821 title image. Before I ran them through an optimizer, they weighed 260 KB; after, 236 KB (using only lossless optimizations). 8% of the total download was dedicated to the content. 5 discrete external companies were informed of my visit to the page and given the opportunity to run artibrary JavaScript on it.

If these are the symptoms, what is the cure? My basic principles are these:

I’ve been building with these principles in mind, and I spent a few hours this optimizing it further. What do the results look like? The heaviest page, the marketing page, today weighs 110 KB with a cold cache, and 4.6 KB warm. A similar page on GitHub.com2 weighs 2900 KB cold, 19.4 KB warm. A more typical page on weighs 56.8 KB cold and 31.9 KB warm, after 2 HTTP requests; on GitHub the same page is 781 KB cold and 57.4 KB warm, 118 requests. This file is 29.1 KB. The overhead is 27.6 KB cold and 2.7 KB warm. The GitHub overhead is respectively 751.9 KB and 28.2 KB. There’s also a 174-pixel-tall ad on GitHub encouraging me to sign up for an account, shown before any of the content.

To be fair, the GitHub page has more features. As far as I can tell, most of these aren’t implemented in the page, though, and are rather links to other pages. Some of the features in the page include a dropdown for filtering branches and tags, popups that show detail when you hover over an avatars, some basic interactivity in the search, all things that I can’t imagine taking up much space. Does this justify an order of magnitude increase in resource usage?

Honestly, GitHub does a pretty good job overall. Compared to our New York Times example, they’re downright great. But they could be doing better, and so could we all. You can build beautiful, interactive websites with HTML and CSS alone, supported by a simple backend. Pushing the complexity of rendering your single-page app into the frontend might save you miniscule amounts of server-side performance, but you’d just be offloading the costs onto your visitor’s phone and sucking their battery dry.

There are easy changes you can make. Enable caching on your web server, with a generous expiry. Use a hash of your resources in the URL so that you can bust the cache when you need to. Enable gzip for text resources, and HTTP/2. Run your images through an optimizer, odds are they can be losslessly compressed. There are harder changes, too. Design your website to be usable without JavaScript, and use small amounts of it to enhance the experience - rather than to be the experience. Use CSS cleverly to provide interactivity3. Find ways to offload work to the server where you can4. Measure your pages to look for places to improve. Challenge yourself to find the simplest way of building the features you want.

And if anyone at Google is reading, you should try recommending these strategies for speeding up pages instead of pushing self-serving faux standards like AMP.

  1. Greater than the vertical resolution of my desktop display. ↩︎

  2. You may have to log out to see this. ↩︎

  3. For example, check out how I implemented the collapsable message details on the archives ↩︎

  4. I did this when I upgraded to Font Awesome 5 recently. They want you to include some JavaScript to make their SVG icons work, but instead I wrote a dozen lines of Python on the backend which gave me a macro to dump the desired SVG directly into the page. ↩︎

Have a comment on one of my posts? Start a discussion in my public inbox by sending an email to ~sircmpwn/ [mailing list etiquette]

Articles from blogs I read Generated by openring

Willingness to look stupid

People frequently1 think that I'm very stupid. I don't find this surprising, since I don't mind if other people think I'm stupid, which means that I don't adjust my behavior to avoid seeming stupid, which results in people thinking tha…

via Dan Luu October 21, 2021

Status update, October 2021

Hi! Another month, another status update. Let’s dig in! The highlight of this month is the launch of, a hosted IRC bouncer service based on soju and gamja. The service is in closed beta for now, feel free to ping me if you want to try it out! It to…

via emersion October 18, 2021

What's cooking on SourceHut? October 2021

Welcome back for another month’s status update! As of today, our community numbers at 25,074 — the first time we’ve had over 25,000 users — after another 522 users joined our ranks. Please show them the courtesey and patience you’ve always done as they learn…

via Blogs on Sourcehut October 15, 2021