Published May 10, 2012
HPC2 Compute Resources Carry Airflow Research Forward
The University at Buffalo's Center for Computational Research cluster is being used by Clarkson researchers Michael Brazell and Brian Helenbrook to help better understand indoor air quality. Numerical simulations of air flow can help improve indoor air quality by illustrating the transport of particulate matter and assisting in the design of more efficient ventilation systems.
Brazell and Helenbrook, along with CCR computational scientist Nate Barlow, simulated a displacement ventilation system, where cold air enters in the bottom of the room and is driven upwards by the thermal plume generated by a hetated mannequin. The mannequin was heated to simulate the normal body temperature of a human. The resulting animation (an iso-contour of temperature colored with contours of velocity magnitude) showed the unsteady behavior of a buoyancy driven flow seen in displacement ventilation systems.
The fluid flow simulation is governed by the compressible Navier-Stokes equations and discretized using an in-house 3rd order accurate finite element method on tetrahedra. The domain contains 500,000 tetrahedra with a total of 3.8 million degrees of freedom and was solved using 144 processors. The Reynolds number is Re=2500, which is based on the inlet velocity and inlet diameter and the Richardson number is Ri=14.6 which is based on the room height and inlet velocity. The research was funded by the Air Pollution Educational and Research Grant Program (APERG) and a CARTI Syracuse Center of Excellence Graduate Fellowship.
216 Intel Xeon E5-2650v2 nodes:
2.6 GHz per core,
8 cores per socket,
2 sockets per node,
8 FLOPS per cycle
333 GFLOPS/node peak
(72 TFLOPS total)
144 nodes with Mellanox FDR IB (56 Gbps)