TFETI Coarse Space Projectors Parallelization Strategies
This paper deals with an analysis of various parallelization strategies for the TFETI algorithm. The data distributions and the most relevant actions are discussed, especially those concerning coarse problem. Being a part of the coarse space projector, it couples all the subdomains computations and accelerates convergence. Its direct solution is more robust but complicates the massively parallel implementation. Presented numerical results confirm high parallel scalability of the coarse problem solution if the dual constraint matrix is distributed and then orthonormalized in parallel. Benchmarks were implemented using PETSc parallelization library and run on HECToR service at EPCC in Edinburgh.
Keywordsdomain decomposition FETI Total FETI TFETI PETSc natural coarse space coarse problem
Unable to display preview. Download preview PDF.
- 3.Balay, S., Gropp, W., McInnes, L.C., Smith, B.: PETSc 3.0.0 Users Manual, Argonne National Laboratory, http://www.mcs.anl.gov/petsc/
- 4.Lingen, F.J.: Efficient Gram-Schmidt orthonormalization on parallel computers. Research report, Department of Aerospace Engineering, Delft University of Technology (1999)Google Scholar
- 8.HECToR Home Page, http://www.hector.ac.uk/