NPRE Prof. Rizwan Uddin is among scientists on the Urbana campus who have gained access to the Blue Waters supercomputer.
Uddin's team has been granted an exploratory award designed to help the team explore the use of the facility for research challenges. His team's project is Scaling up of a Highly Parallel LBM-based Simulation tool (PRATHAM) for Meso- as well as Large-Scale Laminar and Turbulent Flow and Heat Transfer.
“We propose to test and scale PRATHAM – an LBM-based code for 3D, time dependent, laminar and turbulent fluid flow – on Blue Waters,” Uddin said. NPRE alumnus Prashant Jain, MS 06, PhD 10, currently at Oak Ridge National Laboratory, developed PRATHAM with his colleagues at ORNL. Jain is also a team member on this project.
According to Uddin, as the spatial scale of some applications of fluid mechanics and heat transfer continues to drop, researchers are finding that the classical approach of solving the Navier-Stokes equations cannot be applied to an increasing number of micro- and nano-scale applications. Limitations on the parallel scalability of codes that are based on the classical approaches are becoming evident.
While classical approaches to turbulence – such as k-eps, LES, DNS, etc. – in the context of the N-S equations continue to show slow improvement in predictive capabilities, alternate approaches—such as LES with LBM – must be explored because they can take better advantage of advances in computer architecture.
“We propose to perform scalability tests of a laminar and turbulent flow and heat transfer simulation capability to enable high-fidelity, time-dependent, fluid flow and heat transfer analyses of meso-scale (as well as larger scale) systems,” Uddin said. “The simulation tool is based on the Lattice Boltzmann Method (LBM), a relatively novel kinetic approach to simulate transient flow dynamics.
“Because of the LBM’s excellent parallel scalability, the tool is ideally suited for multicore and hybrid GPU-based supercomputing clusters,” he continued. “This tool provides us with an alternative to the existing Navier-Stokes Equation (NSE)–based Large or Very Large Eddy Simulation (LES/VLES) solvers and has been shown to surpass some of their practical limitations.”
The broader sub-goals of this research are
- Implement and test the LBM code on Blue Waters
- Demonstrate the scalability of the code to tens of thousands of processors
- Apply the code to solve a challenging turbulent flow problem in a rod bundle in a nuclear power plant to demonstrate its potential
- Add new modeling capabilities to address multi-physics problems of interest to the university community and funding agencies.
Startup allocation on Blue Waters will help complete the first two tasks; possibly in 12 to 18 months.