Android Phones Can Substitute for Supercomputers

There's an app for almost everything. Now add one that can run calculations from a supercomputer on a Nexus One phone in real time and without the need for internet connectivity.

Researchers at Massachusetts Institute of Technology and Texas Advanced Computing Center have created an Android app that can take simulations from the powerful Ranger supercomputer and solve them further on the mobile phone.

"The idea of using a phone is to show we can take a device with one chip and low power to compute a solution so it comes as close to the one solved on a supercomputer," John Peterson, a research associate at the Texas Advanced Computing Center, told Wired.com.

Many researchers depend heavily on supercomputers capable of millions of calculations per second to simulate problems and advance their studies. Texas Computing Center's Ranger supercomputer went live in 2008 with 62,976 CPU cores, 123 terabytes of memory, 1.73 petabytes of disk space and 579.4 teraflops of performance.

But massive machines such as the Ranger are not easily available. Researchers have to book time on them and they aren't available for computations that need to be done quickly. Supercomputers also can't be carried into field experiments. Having a device in hand that could help solve a problem quickly can be handy.

That's where a technique called "certified reduced basis approximation" comes into play. The method lets researchers take a complex problem, define the values that are most relevant to the problem and set the upper and lower bounds. David Knezevic, a post-doctoral associate at MIT and Anthony Patera, a professor at the school, refined the technique to make it work on a smartphone. They did it by including strong error bounds that show how close they are to an actual supercomputer solution.

"It's demonstrating that with a small processor, you can still get a meaningful answer to a big problem," says Peterson.

The app is just one half of the solution, though. A supercomputer still has to create the reduced model that can be transferred to the phone as an app. When outside the office, researchers can enter values into the app to find answers quickly or visualize data.

For instance, for a problem in fluid dynamics, researchers will spend a day or two simulating a model using a supercomputer like Ranger. Of that computation, they will take a small amount of data and store it on a server as a reduced model.

This reduced model can be used to perform simulations on a cellphone, offering answers near instantaneously for use in real-world applications.

"The payoff for model reduction is large when you can go from an expensive supercomputer solution to a calculation that takes a couple of seconds on a smart phone," Knezevic told a writer at the Texas Advanced Computing Center. "That's a speed up of orders of magnitude."

There's one disadvantage though. The smartphone app has to be customized for the problem it is solving, so it's not universal.

"If a researcher came along with a problem, he would have to code up his own equation within the framework to represent it on the phone," says Peterson. "What he would develop would be specific to the problem."

For now, the researchers have made their app available through files on SourceForge.

Check out their video showing how the app works:

Photo: Texas Advanced Computing Center


Android Phones Can Substitute for SupercomputersWired.com has been expanding the hive mind with technology, science and geek culture news since 1995.