A Brown University team vies for one of HPC’s top honors
By Caitlin Elizabeth Rockett
The National Institute for Computational Sciences’ (NICS) Kraken supercomputer hosted one of the finalists for this year’s Gordon Bell Prize—to be awarded at the SC11 meeting in Seattle—for their innovative multi-scale studies on blood flow in the brain.
Since 1987, computational researchers have been recognized for their contributions to the development of parallel processing via the Association for Computing Machinery’s Gordon Bell Prize. Established and funded by HPC forefather Gordon Bell, the award emphasizes high-performance codes with applied impacts on science.
This year’s prize features two categories: peak performance in terms of flops (floating point operations per second), and special use of innovative techniques on real-world applications. The latter category will include a team lead by Leopold Grinberg, a senior research associate at Brown University. Using Kraken, the group developed models capable of exploring the interactions between red blood cells and investigating the activities that occur within these blood cells.
“One of our main focal points was the progression of aneurisms that can develop in the brain,” explained Grinberg. An aneurysm is an abnormal swelling of a portion of an artery due to weakness in the wall of the blood vessel. They can occur in a number of places in the body and can lead to massive bleeding, clots, strokes, and death if the artery ruptures.
There are treatments that can help to prevent arterial rupture, but the medical community has an incomplete understanding of what causes ruptures, and doctors have limited knowledge about which stage of the aneurysm to apply these techniques. Arterial cells may cause an irregular accumulation of platelets—short-lifespan cells that naturally act to stop bleeding—ultimately producing a blood clot. To study this kind of cellular interaction requires computational models that can address the varying scales of size, from single micrometer-sized cells to centimeter-sized accretions of cells, and Grinberg’s team did just that.
Coupling two codes—NEKTAR and a modified version of LAMMPS—the team was able to simultaneously solve blood flow equations at different scales using over 97,000 of Kraken’s nearly 113,000 cores, painting a realistic picture of blood as it travels through vessels. These simulations needed as many as 1 billion particles to provide an accurate description of cellular interaction.
“That’s a big number,” Grinberg said. “To simulate the interaction between these particles you need a lot of computing power. If you want to do it fast you need a whole lot of computing power, not to mention the necessary memory.” In addition to Kraken, the team also used resources at the Leadership Computing Facilities at Argonne (ANL) and Oak Ridge National Laboratories.
Over 2 years of research on Kraken allowed Grinberg’s team to investigate and evaluate arterial elasticity and increased viscosity of blood in different diseases. The team also learned that certain types of aneurysms produce distinct sounds in the range of 30 Hertz (humans have an auditory range between 20 and 20,000 Hertz) that can be used to pinpoint the condition. Grinberg hopes that with validation of these models, doctors will eventually be able to use computational simulations to determine best-practice methods for inserting stints that can prevent arterial rupture.
Grinberg’s team includes George Karniadakis, also of Brown University; Joseph A. Insley, Vitali Morozov, Michael E. Papka, and Kalyan Kumaran of ANL; and Dmitry Fedosov of the Institute of Complex Systems in Germany. They and other Gordon Bell finalists will present their research during technical sessions at SC, and winners will be announced on November 17 during a special awards presentation.
Whether they win or not, Grinberg’s team is honored to be nominated for such a prestigious award. They have provided unprecedented contributions to both medical and computational sciences, and the greatest reward for their research is the insight they have conveyed and the discoveries they will continue to deliver as capabilities of computational resources grow.