Load Balancing for Mesh Based Multi-Physics Simulations in the Arcane Framework
Céedric Chevalier, Gilles Grospellier, Franck Ledoux, J.-Ch. Weill · Civil-comp proceedings · 2012
Nowadays, industrial applications frequently need to deal with large and complex numerical simulations. To run, they require the use of large computers, such as TERA100 at CEA DAM. To efficiently exploit such architectures, be able to have a correct load balance of the computations is a key issue, as the speed of the overall computation is very often given by the speed of the slowest process. This paper presents load balancing issues for large multi-physics simulations, and specially it is focused on engineering aspects. Using the ARCANE framework, we show how to make an efficient use of state-of-the-art partitioning tools. We describe our methodology and we also discuss what improvements are needed on the path towards peta-scale and exa-scale computing.