Parallelization of an unstructured Navier–Stokes solver using a multi-color ordering method for OpenMP

•A multi-color Gauss–Seidel (MCGS) method is developed for OpenMP parallelization.•Using MCGS, an unstructured-grid Navier–Stokes solver is parallelized.•An algorithm for painting multi-colors is developed for MCGS.•The uniqueness of the solution of MCGS is confirmed with double precision accuracy....

Full description

Saved in:
Bibliographic Details
Published in:Computers & Fluids Vol. 88; pp. 496 - 509
Main Authors: Sato, Yohei, Hino, Takanori, Ohashi, Kunihide
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.12.2013
Elsevier BV
Subjects:
Central processing units
Computer simulation
Mathematical analysis
Mathematical models
Multi color ordering method
Multi grid method
Navier-Stokes equations
OpenMP
Order disorder
Parallel processing
Parallelization
Solvers
Unstructured grid
Unstructured grid
OpenMP
Multi grid method
Multi color ordering method
Parallelization
ISSN:0045-7930, 1879-0747
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•A multi-color Gauss–Seidel (MCGS) method is developed for OpenMP parallelization.•Using MCGS, an unstructured-grid Navier–Stokes solver is parallelized.•An algorithm for painting multi-colors is developed for MCGS.•The uniqueness of the solution of MCGS is confirmed with double precision accuracy. A multi-color ordering method has been developed for the Gauss–Seidel (GS) method in the framework of the unstructured-grid based Navier–Stokes equations solver using OpenMP. The multi-color ordering method is required to avoid the data race condition in do-loop parallelization and to achieve the uniqueness of a solution of GS. A coloring algorithm of painting neighbor cells with different colors is proposed for the multi-color ordering method. The method is tested for four sample simulation cases: one case of two-dimensional simulation and three cases of three-dimensional simulation. Through the sample simulations, the uniqueness of the solution of the Multi-Color ordering Gauss Seidel (MCGS) method is verified, and the convergence ratio of MCGS is found to be in the similar level to that of GS and better than the Jacobi method. The parallel efficiency is examined for workstations with two hexa-core CPUs or two octa-core CPUs. Although the parallel efficiency is dependent on computer systems and simulation cases, the speed up ratio of MCGS reaches 14 using two octa-core CPUs in the maximum case using 14 million cells.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0045-7930
1879-0747
DOI:10.1016/j.compfluid.2013.10.008