SC17 Denver, CO

P07: PORTAGE - A Flexible Conservative Remapping Framework for Modern HPC Architectures


Authors: Rao V. Garimella (Los Alamos National Laboratory), Peter D. Crossman (Los Alamos National Laboratory), Gary A. Dilts (Los Alamos National Laboratory), Rajeev S. Erramilli (Los Alamos National Laboratory), Charles R. Ferenbaugh (Los Alamos National Laboratory), Angela M. Herring (Los Alamos National Laboratory), Eugene Kikinzon (Los Alamos National Laboratory), Chris M. Malone (Los Alamos National Laboratory), Navamita Ray (Los Alamos National Laboratory), Mike L. Rogers (Los Alamos National Laboratory)

Abstract: Portage is a massively parallel remapping framework to transfer fields between general polyhedral meshes while conserving integral quantities of interest. The framework also has the capability to remap data between two point clounds. Portage is templated on the component classes required in conservative remapping - search, intersection and interpolation as well as on the mesh and field managers. Applications supply Portage with custom components while the framework takes care of distributed parallelism using MPI and threaded parallelism using NVIDIA Thrust to scale to many thousands of cores. Moreover, the imposition of a functional design on the components used by Portage makes it very amenable to achieve task parallelism with runtime systems such as Legion. Portage has been tested in 2D/3D for remapping between general polygonal and polyhedral meshes and between point clouds. We present scaling results for distributed (MPI) and on-node parallelism (OpenMP) on LANL's HPC machines.
Award: Best Poster Finalist (BP): no

Poster: pdf
Two-page extended abstract: pdf


Poster Index