A Scalable Parallel Multigrid Solver for Three Dimensional Adaptive Mesh Structural Analysis

A parallel multigrid algorithm for solution of adaptive structural analysis meshes (ParMASA) is described. The user inputs a coarse mesh. The coarse mesh is successively solved, error estimated, and refined. The refinement simultaneously reduces discretization error and creates a hierarchy of meshes for use by the multigrid solver. The mesh may be refined in an arbitrarily irregular anisotropic manner. A block data structure is implemented to facilitate the parallelization of this complex refinement algorithm. For parallel runs an efficient algorithm must have balanced load in all algorithm steps. In addition communication must be minimized. A variety of procedures to optimize parallel performance by modifying the structure of multigrid cycle, smoothers, and communication patterns are discussed. Benefits of the advanced refinement and efficient parallelization are discussed. Good speedups to hundreds of processors are obtained on an SGI Origin 2000. ParMASA shows excellent performance a 256 processor SGI Origin 2000. ParMASA can perform all required IO, refinement, error estimation, and solution steps of an over 10 million degree-of-freedom system of equations in only 142 seconds.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Civil Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Crane, Nathan Karl

Contributors dc:contributor

• Hjelmstad, Keith D.
• Parsons, I.D.

• Computer Science

Language dc:language

eng

Identifier

(MiAaPQ)AAI3070285

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/83194