Associate Professor Steve de Bruyn Kops of the Mechanical and Industrial Engineering Department at the University of Massachusetts Amherst leads one of two teams awarded inaugural Frontier Project grants from the Department of Defense (DoD) High Performance Computing Modernization Program (HPCMP). de Bruyn Kops’ project, entitled Multiscale Interactions in Stratified Turbulence, will use high performance computers to simulate turbulence in fluids strongly affected by gravity.
The objective is to understand very fundamental aspects of fluid turbulence, with the results applied to reducing drag in underwater vehicles, improving sensors systems such as radars, and advancing local-scale weather predictions.
The DoD will provide de Bruyn Kops’ project with high-performance computing resources, including 528 million hours of computing time, storage for several petabytes of data, and technical expertise for managing large data sets. A major goal of the project is to share the data with researchers around the world without moving the data. In addition, DoD is supporting the project with $750,000 over five years.
As de Bruyn Kops explained, turbulent flows are all around us and are always complex. In fact, turbulence is considered to be one of the hardest classical physics problems. When the flows involve light fluid on top of heavy fluid, though, the turbulence becomes even more difficult to understand and predict. Anyone who has flown in an airplane is familiar with the violent and patchy turbulence that occurs in the upper atmosphere. Since the ocean and atmosphere are so complex, simplified simulations are employed to understand what is happening. For research, the simulations must capture large and small eddies over a very wide range of sizes, a process which requires computers with hundreds of terabytes of memory and petabytes of disk storage.
The necessity for high-performance computing is why the DoD Frontier Project provided de Bruyn Kops’ project with about 100 million hours per year on some of the largest DoD research computers. To put it in perspective, using this much time would require running about 11,000 computer processors all day every day. Actually, de Bruyn Kops' team will use up to 132,000 processors in 24-hour bursts. To buy access to this many processors from one of the cloud providers would cost more than $5 million per year. Cloud computers, though, do not have the ultra-high-speed connections between processors that are required for solving turbulence problems and could not actually enable this research. They also do not come with the free expertise provided by the DoD High Performance Computing Modernization Program.
“So this puts the UMass College of Engineering in the top league for high-performance computing,” said de Bruyn Kops.
Thanks largely to this access to high performance computing, de Bruyn Kops’ research will deliver the first numerically-generated, stratified-turbulence data set featuring so-called “high Reynolds numbers” and dynamic ranges relevant to engineering in the ocean and atmosphere. In fluid mechanics, high and low Reynolds numbers are used to characterize different flow regimes, such as laminar or turbulent flow. Laminar flow occurs at low Reynolds numbers, where viscous forces are dominant, and is characterized by smooth, constant fluid motion. Turbulent flow occurs at high Reynolds numbers and is dominated by inertial forces, which tend to produce chaotic eddies, vortices, and other flow instabilities.
Stratified turbulence behaves much differently at low Reynolds numbers than at high Reynolds numbers, and almost all existing data sets for stratified turbulence are at fairly low Reynolds numbers. de Bruyn Kops’ research will provide the most detailed simulation data to date on stratified turbulence at high Reynolds numbers.
Ultimately, understanding turbulence at very high Reynolds number will require accurately measuring it in the ocean and atmosphere. A component of de Bruyn Kops' project is to use the simulated flows to understand assumptions that are made when converting optical measurements of turbulence to turbulence data. Another component of the project is modeling. Since running simulations requiring millions of computer hours is not practical for solving engineering problems, collaborators at the University of Washington and Cornell University will use the large simulations to improve high-speed models for stratified flows that are suitable for designing engineered systems.
Frontier Projects are designed to explore science and technology with high-performance computational approaches that are not otherwise possible. The projects awarded are expected to be among the most computationally demanding projects the DoD will address. This is the first year Frontier Projects have been awarded.
“Frontier Projects are intended to foster the pursuit of DoD problems leveraging the use of multi-disciplinary teams, DoD's largest supercomputers, and advanced computational modeling techniques,” explained John E. West, the director of the DoD High Performance Computing Modernization Program. “A typical Frontier Project will last on the order of 5 years and involve the teaming of HPCMP expertise and supercomputers with DoD laboratory and test center domain expertise. Some projects may also leverage partnerships with academia supporting the laboratories, test centers, and/or the HPCMP.” (August 2013)