VH-VE Implementation for Direct Numerical Simulation of Weakly Compressible Turbulence on SX-Aurora TSUBASA

Mitsuo Yokokawa (Tohoku University)

Abstract

A direct numerical simulation (DNS) code for weakly compressible turbulence, with a Mach number of approximately 0.3, has been developed to investigate the spatial resolution dependence of fine-scale dissipative structures. The simulation targets three-dimensional homogeneous isotropic turbulence in a compressible fluid governed by the continuity equation, the Navier-Stokes equations, and the total energy equation. These governing equations are spatially discretized using an eighth-order compact difference scheme for the convective terms and an eighth-order central difference scheme for the diffusive terms. Time integration is performed using a third-order TVD Runge?Kutta method with a constant time step. Statistical data, including energy spectra, are computed every several dozen time steps during time marching. These calculations are somewhat time-consuming due to the need to transform velocity fields from physical space to spectral space using parallel Fast Fourier Transforms (FFTs). To alleviate this computational overhead, the statistical data processing is offloaded to the Vector Host (VH) of the SX-Aurora TSUBASA system. As a result, the overall computation time is reduced compared to performing all calculations solely on the Vector Engine (VE). In the presentation, we discuss the implementation of the DNS code using the VH-VE multi-architecture model of the SX-Aurora TSUBASA system.

<= Go back