Load balancing angular adaptivity on energy dependent reactor problems

Abstract

This paper presents a novel parallel algorithm capable of solving problems with multiple energy groups, such as reactor core problems, using energy dependent angular $h$-adaptivity coupled with Dynamic Load Balancing (DLB). We use the C5G7 NEA benchmark as an example with our in-house code FETCH2 and compare runtimes and Continuous Degrees of Freedom (CDOFs) between energy independent angular adapting runs with and without DLB. Our results demonstrate a 32% reduction in CDOFs and a 38% reduction in runtime in favour of the energy dependent angular adaptivity runs with DLB when compared to energy independent angular adaptivity.

Publication
PHYSOR2020 conference proceedings