To Open The Sky

In the beginning, the World Wide Web ran on pure HTML — a markup language derived from the more complex SGML. This HTML, or Hypertext Markup Language, was simple enough to learn easily — and to process quickly. The result was that most Web pages loaded quickly, in a second or two. (And when they didn't, the reason was probably the slower speed of Internet connections combined with the use of large images: especially bitmaps. The introduction of the JPEG image format was a boon to Web users.)1

HTML had a problem, though: it handled both content and presentation, or style. This led to highly inconsistent presentations. For example, HTML provides for six levels of headings, H1 through H6. Lower-level headings are displayed in smaller fonts, as befits their lesser importance. But if someone wants a smaller top-level heading and uses H6 for it, this hierarchy is broken for software such as screen readers which must rely on the headers being used in strict order to properly present the page for vision-impaired users.

Soon, CSS (Cascading Style Sheets) arrived to deal with the problem. While they added an extra layer of complication, they did not significantly slow down the performance of the Web. This was a good thing, because even when the Web was new and comparatively lean of gizmos, users complained when a page took longer than a few seconds to fully load. Jacob Nielsen, a prominent usability expert, in 1997 laid down a standard: pages should load within one second, on average, to avoid user frustration.2

Contrast that with the situation we have today, twenty years later. Most pages take tens of seconds to load completely. This is not due to hardware speeds; computer clock speeds have been boosted by a factor of more than 10,000 since 1976, and other design changes have led to increased throughput.3 By my rough estimate, bandwidth went from 1,200 baud when the Web started to get popular (circa 1995) to typical bandwidths of 50 Megabits/second by today.4 Rather, it is due to the piling on of interconnections to multiple servers for analytics and mostly for advertising.5

A year ago (3 January 2018), I performed a thoroughly unscientific test of the times (in seconds) a randomly selected group of Web pages take to load.

It's clear that much of this change is due to the need for Web sites to support themselves through advertising. I don't object to advertising per se — but when it slows down my browsing well beyond the point of frustration, or messes with my view of the site so that I'm constantly having to adjust the elevator bars to read the story smoothly, I call foul.