Center for Process Simulation and Design, University of Illinois Robert B. Haber, Duane D. Johnson, Jonathan A. Dantzig, DMR-0121695 Scalable Parallelization of Spacetime Discontinuous Galerkin Method in ParFUM L. Kale, S. Chakravorty, T. Wilmarth, N.

1
Center for Process Simulation and Design,
University of IllinoisRobert B. Haber, Duane D.
Johnson, Jonathan A. Dantzig, DMR-0121695Scalabl
e Parallelization of Spacetime Discontinuous
Galerkin Method in ParFUML. Kale, S.
Chakravorty, T. Wilmarth, N. Choudhury, I. Doole,
University of IllinoisDepartment of Computer
Science
Research objectives The Spacetime Discontinuous
Galerkin (SDG) method provides a powerful means
for simulating shock wave propagation and for
coupling continuum and atomistic models. SDG
methods nest local finite element solutions
within an advancing-front mesh generation
algorithm. This sub-project seeks efficient
patch-wise-parallel mesh generation and
solution. Approach In previous parallelization
of SDG methods, a master process determined mesh
patches to solve and sent them to slave
patch-solving processes. This approach had
limited scalability due to communication
bottlenecks at the master process. ParFUM, the
Charm Parallel Framework for Unstructured
Meshes, enables an approach that distributes the
space-mesh amongst processors. Each portion of
the space mesh generates its own spacetime
patches to solve in parallel. ParFUM handles
communication, synchronization, locking and
ghosting to provide uniform access to both local
and non-local portions of the mesh. Significant
results We demonstrate scalable parallel
performance for non-adaptive problems up to 256
processors. ParFUM now supports parallel adaptive
mesh modifications and work is underway to
integrate adaptive SDG methods with the new
parallel approach. Broader impact Efficient
parallel adaptive mesh generation for
time-dependent problems with an integrated
solution procedure has the potential to benefit
an extensive set of applications in science and
engineering.