Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
756731 | Computers & Fluids | 2013 | 9 Pages |
As high performance computing (HPC) systems become more accessible, numerical simulations of various physical systems has become a popular research tool. These methods usually require the solution of a large system of coupled equations which, to take advantage of parallel processing, are usually solved by decomposing the domain. A particularly challenging problem in the area of computational fluid dynamics (CFD) is free surface phenomena which requires a moving domain and with the inclusion of the relevant dynamic boundary conditions, thin boundary layers on the rigid bottom boundary and free surface, which scale with the Reynolds number as Re−1/2. To capture the flow in near wall regions at higher Reynolds numbers requires the solution of ever larger systems of equations. The implementation of these CFD codes to run efficiently on HPC systems is not trivial. This work couples together various computational techniques to be able to simulate high Reynolds number two dimensional free surface flows. This is accomplished through a simple manual domain decomposition technique and by linking to external high performance libraries for parallel vectors/matrices and system solution. The code was tested on UCL’s supercomputer Legion and demonstrated linear weak scaling in tests carried out with up to 60 m nodes and up to 256 cores. The laminar boundary layer under a solitary wave was investigated to highlight the codes ability to resolve the thin boundary layers present. The technique was then applied to analyse sediment transport under a solitary wave.
► Various numerical techniques are presented to carry out free surface simulations. ► Discussion of solvers and preconditioners for the solution of system of equations. ► Linear weak scaling is shown across 256 cores. ► The laminar boundary layer under a solitary wave is studied to validate the method. ► Sediment transport under a solitary wave is studied.