Computational Performance of a LES Solver for Supersonic Jet Flow Applications

TL;DR

This paper evaluates the computational performance and scalability of an in-house LES solver for simulating high-fidelity supersonic jet flows across various mesh sizes using high-performance computing resources.

Contribution

It introduces a parallelized LES solver with detailed performance analysis on large meshes up to 1 billion points, demonstrating its strong scalability on high-performance clusters.

Findings

Achieved efficient parallel performance up to 400 processors.

Demonstrated scalability across meshes from 5.9 million to 1 billion points.

Provided speedup and efficiency metrics for high-fidelity jet flow simulations.

Abstract

An in-house large eddy simulation tool is developed in order to reproduce high fidelity results of compressible jet flows. The large eddy simulation formulation is written using the finite difference approach, with an explicit time integration and using a second order spatial discretization. The energy equation is carefully discretized in order to model the energy equation of the filtered Navier-Stokes formulation. Such nu\-me\-ri\-cal studies are very expensive and demand high performance computing. Message passage interface protocols are implemented into the code in order to perform parallel computations. The present work addresses the computational performance of the solver running on up to 400 processors in parallel. Different mesh configurations, whose size varies from approximately 5.9 million points to approximately 1.0 billion points, are evaluate in the current paper. Speedup and efficiency curves are evaluated in order to assess the strong scalability of the solver.