| The purpose of this research is to investigate and improve parallel discrete ordinates (S;Spatial domain decomposition schemes result in low parallel speedups and efficiencies due to two main effects: high convergence penalties and load imbalance. For standard parallel iterative techniques such as Jacobi (where the fluxes at subdomain boundaries are previous iterates), the convergence rate is slow. The load imbalance is inherent to the S;A new Multi-Color iterative scheme, where subdomains are grouped into colors and accordingly executed in parallel by colors, is developed and implemented on CRAY C90 and J90 shared-memory machines. This scheme results in significantly lower iteration ratios for high numbers of subdomains (as low as 1.32 for 48 subdomains with 12 colors), and accordingly higher speedups and efficiencies.;The directional ;Two acceleration schemes, coarse mesh rebalance (CMR) and diffusion synthetic acceleration (DSA), and their effects on parallel transport calculations are compared. The CMR calculations comprise a small fraction (... |
