Trias, F. X.; Gorobets, A.; Soria, M.; Oliva, A.
Lecture notes in computational science and engineering
Vol. 67, p. 267-274
DOI: 10.1007/978-3-540-92744-0_33
Data de publicació: 2009
Article en revista
A code for the direct numerical simulation (DNS) of incompressible turbulent flows that provides a fairly good scalability for a wide range of computer architectures has been developed. The spatial discretization of the incompressible Navier-Stokes equations is carried out using a fourth-order symmetry-preserving discretization. Since the code is fully explicit, from a parallel point of view, the main bottleneck is the Poisson equation. In the previous version of the code, that was conceived for low cost PC clusters with poor network performance, a Direct Schur-Fourier Decomposition (DSFD) algorithm was used to solve the Poisson equation. Such method, that was very efficient for PC clusters, can not be efficiently used with an arbitrarily large number of processors, mainly due to the RAM requirements (that grows with the number of processors). To do so, a new version of the solver, named Krylov-Schur-Fourier Decomposition (KSFD), is presented here. Basically, it is based on the Direct Schur Decomposition (DSD) algorithm that is used as a preconditioner for a Krylov method (CG) after Fourier decomposition. Benchmark results illustrating the robustness and scalability of the method on the MareNostrum supercomputer are presented and discussed. Finally, illustrative DNS simulations of wall-bounded turbulent flows are also presented.
A code for the direct numerical simulation (DNS) of incompressible turbulent flows that provides a fairly good scalability for a wide range of computer architectures has been developed. The spatial discretization of the incompressible Navier-Stokes equations is carried out using a fourth-order symmetry-preserving discretization. Since the code is fully explicit, from a parallel point of view, the main bottleneck is the Poisson equation. In the previous version of the code, that was conceived for low cost PC clusters with poor network performance, a Direct Schur-Fourier Decomposition (DSFD) algorithm was used to solve the Poisson equation. Such method, that was very efficient for PC clusters, can not be efficiently used with an arbitrarily large number of processors, mainly due to the RAM requirements (that grows with the number of processors). To do so, a new version of the solver, named Krylov-Schur-Fourier Decomposition (KSFD), is presented here. Basically, it is based on the Direct Schur Decomposition (DSD) algorithm that is used as a preconditioner for a Krylov method (CG) after Fourier decomposition. Benchmark results illustrating the robustness and scalability of the method on the MareNostrum supercomputer are presented and discussed. Finally, illustrative DNS simulations of wall-bounded turbulent flows are also presented.