By Steve Hamm
Beware the pistol shrimp. It stuns small sea creatures with a gun-like claw that fires powerful clouds of bubbles at its prey. The scientific principle that gives the pistol shrimp its mini-superhero powers could also prove valuable to humans–in uses ranging from improving the designs of propellers to helping doctors destroy kidney stones and cancerous tumors. A global collaboration involving IBM scientists, researchers at two European universities and the US Lawrence Livermore National Laboratory could help accelerate the journey of this science into the marketplace.
The multi-disciplinary team used one of the world’s fastest supercomputers to simulate the behavior of clouds of bursting bubbles–handling the highly-complex fluid dynamics problem in a way that was extremely efficient. In the process, they set a new record in supercomputing in fluid dynamics and, as a result, the team on Nov. 21 won the coveted Gordon Bell Prize from the Association for Computing Machinery.
Alessandro Curioni, head of mathematical and computational sciences at IBM Research – Zurich, described the adrenaline rush of working on the project. The team ran into one problem after another, and it required a diverse set of skills to solve them. The excitement peaked last April when the team–working around the clock for one week–demonstrated their breakthrough on Lawrence Livermore’s Sequoia computer. Scattered over half the globe, they kept in touch constantly with email, telephones and Skype. “A single group could not have accomplished this. We needed a wide variety of skills. It’s a great example of open collaboration,” he says.Here’s the pistol shrimp in action:
IBM has been collaborating with nearby ETH Zurich for years on a wide range of projects. Two years ago, they set out to show that they could perform the complex bubble simulation on a very large supercomputer. IBM brought skills in algorithmic engineering. The ETH team headed by Petros Koumoutaskos renown in computational sciences . They marshaled forces and reached out to a group at the Technical University of Munich that has deep experience in the cavitation of bubbles. Lawrence Livermore offered up the IBM-designed Sequoia machine–96 racks of clustered computing systems comprised of 1.6 million processor cores capable of a peak performance of 20 petaflops of computation, or 20 quadrillion operations per second.
Until now, scientists faced extreme difficulty in running complex fluid dynamics simulations on large supercomputers. They were only able to use 10% of peak capacity. That’s a problem. It takes a lot of time and electrical energy to run the programs–so it’s too costly. By developing new algorithms and taking advantage of Sequoia’s Blue Gene/Q architecture, the team was able to push capacity utilization to 55% in April and 73% in October. It was a massive fine-tuning project involving a lot of trial and error.
Six members of the team were able to make the trip to the SC13 conference in Denver last week to receive the Gordon Bell Prize in person. Alessandro says the collaboration had been so intense that by the time he met his university and LLNL counterparts face to face, he felt like he had known them since childhood. “You start. You have a dream. You know the steps. But, for an application this big, you never know what will happen. It was an incredible experience,” he says.
In addition to Alessandro, Costas and Petros, the other core members of the research team included Diego Rossinelli, Babak Hejazialhosseini, and Panagiotis Hadjidoukas from ETH Zurich; Adam Bertsch and Scott Futral from LLNL; and Steffen Schmidt and Nikolaus Adams from Technical University of Munich.