Validation and scalability of an open source parallel flow solver. (14th September 2017)

• Record Type: Journal Article
• Authors: Bricteux, L.; Zeoli, S.; Bourgeois, N.
• Is Part Of: Concurrency and computation. Volume 29:Number 21(2017)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2017-09-14
• Subjects: DNS -- OpenFOAM -- finite volume method -- Taylor Green Vortex -- validation -- scalability; Parallel processing (Electronic computers) -- Periodicals; Parallel computers -- Periodicals; 004.35
• DOI: 10.1002/cpe.4330
• Languages: English
• ISSNs: 1532-0626

Abstract

Summary: This article concerns validation and verification (V&V) of the open source OpenFOAM incompressible flow solver that is widely used in both research and industry for the simulation of turbulent flows. Its numerical discretization scheme relies on classical finite volumes that are second order in space on regular grids. The solver is benchmarked using a challenging test case for which direct numerical simulation (DNS) reference data exist. The geometrical configuration of this flow is very simple, yet the behavior of the flow is 3D, complex, and unsteady, requiring an accurate flow solver to capture the associated physics. The physical problem addressed here requires huge computational grids. Computations are thus carried using parallel domain decomposition methods on a fairly large number of cores. The validation study performed here is particularly severe as it consists in comparing OpenFOAM results with DNS data obtained using high‐order codes. As simulations of turbulent flows require great computational resources, results concerning strong and weak scalability of the OpenFOAM flow solver are also presented. The performances of the code are assessed on two different HPC infrastructures. The experiences carried here show that OpenFOAM scales fairly up to 1000 cores on a Tier‐1cluster. The influences of the processor hardware and network topology are also highlighted. This study also provides profiling diagnostics thanks to the Integrated Performance Monitor (IPM).