Araz Eghbal

Space-Time Parallelization of Fluid Dynamics Simulations

• Date: May 16, 2018
• Time: 1:00pm
• Location: ADI Studio, Head Hall

Abstract

Advances in computer hardware technology have made powerful computers more accessible to scientists and engineers. In order to be able to use the computational power of these new architectures parallel computing algorithms need to be improved and updated frequently.

Conventional methods to solve initial value evolution problems use time stepping techniques which involve solving a boundary value problem in each time-step and then sequentially propagating the fi to the next step. Com- mon parallel computing methods are applied to the spatial boundary value portion of the problem which can reach saturation before being able to employ all the available computational power. The parareal algorithm is one of the strongest candidates in the search for parallelization of the temporal dimension of time dependent problems.

In this thesis parareal temporal parallelization framework has been added to the general purpose computational fluid dynamics software EXN/Aero with the goal of testing the feasibility of the method in different types of fluid dynamic applications. New strategies such as filtering of grid dependent fields weak multistage convergence techniques and conservative agglomeration methods are introduced to facilitate the space-time parallelism in these problems.

Approximately 2-3 fold speedup in the computation of select cases is reported. Concerns regarding numerical stability of the methodology when applied to highly nonlinear problems are stressed and CFD solution methods that may benefit the most from this kind of parallel computing technique are mentioned.