Parallelization of an Adaptive Vlasov Solver
Abstract
This paper presents an efficient parallel implementation of a Vlasov solver. Our implementation is based on an adaptive numerical scheme of resolution. The underlying numerical method uses a dyadic mesh which is particularly well suited to manage data locality. We have developed an adapted data distribution pattern based on a division of the computational domain into regions and integrated a load balancing mechanism which periodically redefines regions to follow the evolution of the adaptive mesh. Experimental results show the good efficiency of our code and confirm the adequacy of our implementation choices. This work is a part of the CALVI project.