Basic HTML version of Foils prepared 15 March 1996

Foil 11 Basic Formulae for Hyperplane Parallelization of LU NAS Benchmark

From Further PDE Solvers for the NAS Benchmarks CPSP713 Case studies in Computational Science -- Spring Semester 1996. by Geoffrey C. Fox
1 Essential idea is that we divide equations into sweeps labelled by m
  • m = i + j + k
2 Once one has solved sweeps 1 through m-1, ALL equations of sweep m can be solved INDEPENDENTLY and hence with natural parallelization
3 We will use Owner-Compute strategy and assign each grid-point (as normal) to a given processor
4 Several choices for assignments of grid points to processors and not clear that RNR-93-007 chooses best strastegy
5 Natural Strategy is NOT BLOCK ( divide grid points into P contigous cubic domains) but rather BLOCK-CYCLIC
  • BLOCK decomposition has load-balancing problems
  • BLOCK-CYCLIC has better load-balancing but more communication
  • P denotes number of Processors
in Table To:

Northeast Parallel Architectures Center, Syracuse University, npac@npac.syr.edu

If you have any comments about this server, send e-mail to webmaster@npac.syr.edu.

Page produced by wwwfoil on Sun Feb 22 1998