0100482625

open access

Accepted Manuscript

Performance evaluation of parallel direct numerical simulation code on supercomputer SX-Aurora TSUBASA

Performance evaluation of a parallel DNS code on the supercomputer SX-Aurora TSUBASA

English (英語)

Computers & Fluids

261

105913

Elsevier

2023-07-15

2025-05-15

Although direct numerical simulations (DNSs) of incompressible turbulent flows have been performed since the late 1960s, simulations that reproduce the strongly nonlinear turbulent flows observed in the real world have not been realized. A DNS code is parallelized using pencil decomposition and optimized for the vector-type supercomputer SX-Aurora TSUBASA. A loop blocking technique is used to optimize the code. It is found that some loops of the three-dimensional fast Fourier transform have an optimal block size for loop blocking that enhances the DNS code performance. The optimized DNS code achieves a 3.3-fold speedup compared to the original code. In addition, a computation time estimation model for large-scale DNSs is constructed based on the measured throughput data for all-to-all communications on SX-Aurora TSUBASA. One step of the Runge–Kutta–Gill integration takes one minute to complete for a DNS with 18,432³ grid points on 16,384 vector engines of SX-Aurora TSUBASA.

Loop blocking

学術雑誌論文

This manuscript version is made available under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International license.