Internet photo
SHARE

The worst part of navigating the internet is the wait. While you stare at a white screen, your computer (or let’s be real, your phone) is chasing down thousands of little snippets of code, which point to other pieces of code, over and over again, through ad trackers and Javascript libraries.

Making these load times faster is as simple as building a map, according to new research by MIT’s Computer Science and Artificial Intelligence Lab and Harvard.

When websites load, the computer is shown paths to all kinds of files to download. Different fonts, the Javascript to make things jump around the page, images, and the formatting of the page all usually come from different places. But when these also point to outside files, all that downloading takes time.

MIT and Harvard’s system, called Polaris, makes a map of all these secondary and tertiary download locations. Then the browser can overlap downloads more effectively, saving up to 34 percent of load time. Systems that track these nested objects, called dependencies, already exist, but MIT says theirs is different because it functions regardless of the browser. This is because it’s built in Javascript, something nearly all functional web browsers can run.

The compared hierarchy of page loading
Black lines show the typical order in which objects on a web page are loaded, from top to bottom. The red lines show how Polaris connects the dots. MIT CSAIL/ Harvard

James Mickens, a Harvard professor who’s visiting at MIT, explains the system in terms of a travelling businessperson. If there are 20 cities on the agenda, it’s far more efficient to plan travel from one coast to another, rather than unnecessarily flying back and forth from New York to Los Angeles to Washington D.C. to San Francisco. The Polaris system works in much the same way, making a graph of which objects can be loaded in tandem.

Researchers tested their system on sites that see lots of traffic, like The New York Times and ESPN. Polaris is able to work well on nearly any site because it just organizes the provided content more efficiently.

The researchers are still looking at next steps, according to a representative from MIT, but are interested in feedback on the software and having others iterate on it. They’re planning to make the code public in the coming months.