Opera developer with Address bar speculative prerenderer

December 5th, 2016

In first Opera 43 developer blog post we announced new feature coming in: Address bar speculative prerenderer. Here are some more details:

When you browse the web a lot of time is spent waiting for pages to load and a lot of energy has gone into optimizing page load, both from browser makers and from site makers. Today that is highly optimized and it is hard to make big improvements.

So let us think outside the box, what is the best way to complete a task earlier? To start earlier! And that is the whole trick in the “prerendering” feature.

Predict

There are two ways we can predict what page the user will soon load. When the current page tells us so, and when we can determine from the users actions that they are about to load something.

Pages can use the < link rel=prerender … > tag, and for instance Google uses that for search results if they are pretty sure of what you will load next.

When someone writes in the address bar they are humanly slow. Sometimes it is obvious what they will write after just 1-2 characters but they will just keep writing or arrowing through suggestions for millions or billions of wasted clock cycles.

Result

We expect this feature to results in an average of 1 second faster loads from the address bar. That means that instant page loads become much more common. Of course this depends a lot on how predictable the user is and how fast the user is. I expect you who read this to benefit less than the average user, but I might be wrong.

Security/Privacy

We try hard to not load anything that would be bad to load. Should you notice anything else, let us know right away!

There are also settings in the privacy section to disable this completely.

Monitor

If you want to see how predictable you are, fire up Opera, browse some, then load opera:predictors and see what green lines there are (that is when we “know” what you are about to do). You can also look at the prerender section in opera:net-internals and see when we are prerendering right now and historically.

Limitations

Some studies indicate about 80% of prerendered pages become used, which means that 20% of the prerender page loads are for nothing and has wasted electrons. Therefore we don’t do this in battery saving mode (could argue that we should still do it).

Since every preload (currently) needs to spawn a new process, it also won’t work if you have reached the process limit. Or if you are not predictable enough in your use of the address bar.