Shared-memory, distributed-memory, and mixed-mode parallelisation of a CFD simulation code

Abstract

This paper presents some different approaches to the parallelisation of a harmonic balance Navier-Stokes solver for unsteady aerodynamics. Such simulation codes can require very large amounts of computational resource for realistic simulations, and therefore can benefit significantly from parallelisation. The simulation code addressed in this paper can undertake different modes of aerodynamic simulation and includes both harmonic balance and time domain solvers. These different modes have performance characteristics which can affect any potential parallelisation, as can the specifics of the problem being simulated. Therefore, three different techniques have been used for the parallelisation, shared-memory, distributed-memory, and a combination of the two--a hybrid or mixed-mode parallelisation. These different techniques attempt to address the different performance requirements associated with the types of simulation the code can be used for and provide the level of computational resources required for significant simulation problems. We discuss the different parallelisations and the performance they exhibit on a range of computational resources.