A GPU Computing Platform for Modeling Fluid-Sheared Granular Beds
Abstract
Major Goals: The transport of sediment by flowing water is a fundamental physical process with broad applications in the geological sciences. For example, the ability to predict and control the erosion of granular beds could be used to promote or mitigate erosion, often with significant economic and humanitarian impacts. The process of sediment transport involves the nontrivial coupling of turbulent fluid flow over a rough boundary with the dynamics of granular materials, each of which is difficult to characterize on its own. Thus, a precise determination of the conditions whereby grains first start to move remains an open question. Historically, most approaches have emphasized the role of fluid mechanics, treating the grains not as individual particles but as an averaged statistical system. However, recent research advances have allowed more detailed investigations of the role played by each grain individually. Computer simulations that model the interactions between every pair of grains in a bed of sediment are now feasible. In this project, we aim to utilize graphics processing units (GPUs) to study both thestatistical mechanics of static granular beds, as well as the dynamics of mobile grains, via powerful simulations that can track the behavior of millions of particles at a time. Our simulations of sediment transport will require significant computational resources, and the use of parallelcomputing will be essential in completing them. Until recently, this was done by running the simulations on central processing units (CPUs) that shared data between them via a message passing interface (MPI). Recently, however, a problem with using MPI on CPUs has emerged: the processor clock speeds can no longer be increased because doing so would generate so much heat that the processors would suffer frequent failures.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 30, 2019
- Accession Number
- AD1098071
Entities
People
- Andrew Sherman
- Corey S O'Hern
Organizations
- Yale University