A custom app to interface with a supercomputer and complete some of its calculations? You’ve got it, courtesy of MIT and Texas Advanced Supercomputer Center. And it’s not Apple’s glory this time, it’s Android’s.
Before you get too excited by the thought of all this, it’s not quite as swizzily clever as hooking a boatload of PS2s or PS3s together to form a low-cost supercomputer. But it is clever, and it does involve finishing supercomputer tasks on the diminutive CPU chugging away inside an Android phone.
It’s a combined effort from thinkers at MIT and Texas Advanced Computing Center, and has resulted in a dedicated Android app that lets researchers plan their ongoing computational tasks running on the TAC’s Ranger supercomputer–a 504 teraflop monster with 63,000 CPU cores that’s used on an open basis for scientific research (unlike most supercomputers, which run super-secret military simulations, or purr away inside meteorological offices).
Since supercomputer time has to be carefully rationed out, to meet the demands of each unit’s numerous users, and to make cost-effective use of every second of its uptime, running calculations on one isn’t necessarily a straightforward task (it rarely was when I used one, as an example). For example, before you can formulate a problem to run on the machine, you need to work out what your problem is and code the software carefully–you often can’t easily book the machine to test your code for a short spell, and running urgent calculations isn’t easy either.
Hence some lateral thinking to utilize the tiny CPUs inside the smartphones that so many of us carry around now. The idea uses “certified reduced basis approximations” as a way to quickly and efficiently get a mathematical “feel” for the solution that the full-sized Ranger supercomputer will come up with you face it with a complete task. The bigger, complex task is simplified down, removing some of the mathematical frills and boiling it down to its relevant values and the upper and lower bounds of the solution that the supercomputer will be tasked with–this happens in a short run on the Ranger, and then the code is shot to the Android app on a researcher’s machine. Then, without the need for Net access and whenever inspiration strikes (typically at 3 a.m. when you can’t sleep) the scientist can tap in some test values and get a quick result from the app–including a measure of how close this might be to an actual supercomputer solution.
To keep up with this news, follow me, Kit Eaton, on Twitter.