This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.
Permalink
https://hdl.handle.net/2142/125706
Description
Title
Exascale finite/spectral element simulations
Author(s)
Ratnayaka Mudiyanselage, Thilina Bandara
Issue Date
2024-07-09
Director of Research (if dissertation) or Advisor (if thesis)
Fischer, Paul
Olson, Luke
Kloeckner, Andreas
Rowe, Kris
Doctoral Committee Chair(s)
Fischer, Paul
Department of Study
Siebel Computing &DataScience
Discipline
Computer Science
Degree Granting Institution
University of Illinois at Urbana-Champaign
Degree Name
Ph.D.
Degree Level
Dissertation
Keyword(s)
HPC
GPU
MPI
Graph Partitioning
Schwarz Methods
Coarse-grid
p-Multigrid
Loopy
Programming Models
Clang
Exascale Computing
Abstract
In this thesis, we present contributions by author and his collaborators to improve the efficiency of high-order spectral/finite element simulations (SEM/FEM) on state-of-the-art exascale and pre-exascale supercomputers with general purpose graphics processing units (GPUs). We briefly discuss the exascale computing landscape and the challenges involved in running high-order SEM/FEM simulations on these machines. Then we present contributions which span four important aspects of high-order SEM/FEM simulations. Firstly, we present representative benchmarks for SEM/FEM applications and use those to benchmark new GPU based exascale systems in order to understand their scalability characteristics. Secondly, we present a parallel partitioner suitable for the exascale era that is characterized by large local problem sizes. Our partitioner is based on recursive spectral bisection and is able to partition a mesh in parallel while preserving load balance and reducing mes- sage volume between partitions. Thirdly, we address one of the main bottlenecks in the SEM/FEM simulations: the coarse-grid solve of the p−Multigrid preconditioner used for solving the pressure Poisson equation. We introduce a novel two-level Schwarz based coarse solver which has the potential to scale well on the exascale machines due to its low communication and increased parallelism. Finally, we present a domain specific compiler framework to improve the productivity of the application engineers who use GPU-based supercomputers by reusing domain specific optimization patterns.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.
