Analysis of Massively Parallel Discrete-Ordinates Transport Sweep Algorithms with Collisions
We present theoretical scaling models for a variety of discrete-ordinates sweep algorithms. In these models, we pay particular attention to the way each algorithm handles collisions. A collision is defined as a processor having multiple angles with ready to be swept during one stage of the sweep. The models also take into account how subdomains are assigned to processors and how angles are grouped during the sweep. We describe a data driven algorithm that resolves collisions efficiently during the sweep as well as other algorithms that have been designed to avoid collisions completely. Our models are validated using the ARGES and AMTRAN transport codes. We then use the models to study and predict scaling trends in all of the sweep algorithms.
- Research Organization:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 952433
- Report Number(s):
- LLNL-CONF-407968; TRN: US0902491
- Resource Relation:
- Conference: Presented at: International Conference on Mathematics, Computational Methods, and Reactor Physics, Saratoga Springs, NY, United States, May 03 - May 07, 2009
- Country of Publication:
- United States
- Language:
- English
Similar Records
Denovo--A New Three-Dimensional Parallel Discrete Ordinates Code in SCALE
Parallel S{sub n} Sweeps on Unstructured Grids: Algorithms for Prioritization, Grid Partitioning, and Cycle Detection