A hybrid MPI-OpenMP parallel implementation for pseudospectral simulations with application to Taylor–Couette flow. (5th January 2015)

• Record Type: Journal Article
• Authors: Shi, Liang; Rampp, Markus; Hof, Björn; Avila, Marc
• Is Part Of: Computers & fluids. Volume 106(2015)
• Page Start: 1
• Page End: 11
• Publication Date: 2015-01-05
• Subjects: Direct numerical simulation -- Taylor–Couette flow -- Hybrid parallelization -- Pseudospectral method -- Finite difference; Fluid dynamics -- Data processing -- Periodicals; 532.050285
• DOI: 10.1016/j.compfluid.2014.09.021
• Languages: English
• ISSNs: 0045-7930

Abstract

Highlights: A highly-efficient direct-numerical-simulation method for Taylor–Couette flow is presented. The hybrid MPI-OpenMP parallelization strategy reduces internode communications. The code demonstrates excellent scalability up to O ( 10 4 ) processor cores. The code enables to simulate turbulent Taylor–Couette flow at Reynolds number ∼ O ( 10 5 ) . The code may potentially resolve the controversy of turbulence in Keplerian flows. Abstract: A hybrid-parallel direct-numerical-simulation method with application to turbulent Taylor–Couette flow is presented. The Navier–Stokes equations are discretized in cylindrical coordinates with the spectral Fourier–Galerkin method in the axial and azimuthal directions, and high-order finite differences in the radial direction. Time is advanced by a second-order, semi-implicit projection scheme, which requires the solution of five Helmholtz/Poisson equations, avoids staggered grids and renders very small slip velocities. Nonlinear terms are evaluated with the pseudospectral method. The code is parallelized using a hybrid MPI-OpenMP strategy, which, compared with a flat MPI parallelization, is simpler to implement, allows to reduce inter-node communications and MPI overhead that become relevant at high processor-core counts, and helps to contain the memory footprint. A strong scaling study shows that the hybrid code maintains scalability up to more than 20, 000 processor cores and thus allows to perform simulations at higher resolutions than previously feasible. In particular, it opens up the possibility to simulate turbulent Taylor–Couette flows at Reynolds numbers up to O ( 10 5 ) . This enables to probe hydrodynamic turbulence in Keplerian flows in experimentally relevant regimes.