How ‘bullet time’ will revolutionize exascale computing

The filming technique used in The Matrix will change the way we access the huge computer simulations of the future, say computer scientists
February 12, 2013

( Credit: Warner Bros. Pictures)

The exascale computing era is almost upon us and computer scientists are already running into difficulties. 1 exaflop is 10^18 floating point operations per second, that’s a thousand petaflops. The current trajectory of computer science should produce this kind of  capability by 2018 or so.

How do humans access and make sense of the exascale data sets?

The answer, of course, is to find some way to compress the output data without losing its essential features. Akira Kageyama and Tomoki Yamada from Kobe University in Japan put forward a creative solution: “bullet time”, the Hollywood filming technique made famous by movies like The Matrix, MIT Technology Review reports.

Bullet time is a special effect that slows down ordinary events while the camera angle changes as if it were flying around the action at normal speed.  The technique involves plotting the trajectory of the camera in advance and then placing many high speed cameras along this route. All these cameras then film the action as it occurs.

This footage is later edited together to look as if the camera position has moved. And because the cameras are all high speed, the footage can be slowed down. The results are impressive, as anyone who has seen the Matrix movies or played the video games can attest.

Kageyama and Yamada say the same technique could revolutionize the way humans access exascale computer simulations. Their idea is to surround the simulated action with thousands, or even millions, of virtual cameras that all record the action as it occurs.

Humans can later “fly” through the action by switching from one camera angle to the next, just like bullet time.

Kageyama and Yamada say that the footage from a single camera can be compressed into a file of say 10 megabytes. So even if there are a million cameras recording the action, the total amount of data they produce is of the order of 10 terabytes. That’s tiny compared to the exascale size of the simulation.