Implementation of parallel mesh partition and ghost generation for the Finite Element Mesh framework
Authors:
Sayantan Chakravorty
Parallel Programming Laboratory, Department of Computer Science, University
of Illinois at Urbana-Champaign
Master's Thesis, Department of Computer Science, University of Illinois 2005
The Finite Element Method framework allows the user to develop scalable parallel finite element applications easily. During initialization it reads in an input mesh and partitions it into a large number of chunks that are distributed among different processors. This partition process is sequential and memory intensive. Thus the partition algorithm is a bottleneck for running finite element applications with large meshes on a large number of processors. This thesis parallelizes the partition algorithm to remove the memory bottleneck on one processor. It also presents an algorithm to generate ghost elements in parallel.