Abstract of NEMOFFT Project

Global gyrokinetic simulations of turbulent transport are very demanding in terms of HPC resources, and rely on scalable parallel algorithms. Recent progresses in the inclusion of more physical effects in these codes imply that this demand is expected to increase strongly in the near future. Past parallel scalability studies performed on massively parallel platforms have revealed bottlenecks to the scalability of existing algorithms, especially when large physical system sizes are considered. This proposal focuses on the ORB5 code, in its version (NEMORB) that includes full electron dynamics and electromagnetic effects. It aims at investigating possibilities to remove, or at least to enlarge, some of the current bottlenecks, in order to make global electromagnetic gyrokinetic simulations of turbulence in ITER-size plasmas feasible. The proposed work will investigate a better way to perform the required Fourier transforms in order to assemble the perturbed quantities which, in the present version, requires large amounts of grid data to be transposed across the processors. The present proposal will also look into generic parallel data transpose and Fourier transform issues. These are fundamental kernel problems used not only in NEMORB, but also in several other codes of the EU fusion programme, hence the potential benefit goes beyond the particular application considered in this proposal.