Radiation transport calculations on unstructured grids using a spatially decomposed and threaded algorithm
We consider the solution of time-dependent, energy-dependent, discrete ordinates, and nonlinear radiative transfer problems on three-dimensional unstructured spatial grids. We discuss the solution of this class of transport problems, using the code TETON, on large distributed-memory multinode computers having multiple processors per ''node'' (e.g. the IBM-SP). We discuss the use of both spatial decomposition using message passing between ''nodes'' and a threading algorithm in angle on each ''node''. We present timing studies to show how this algorithm scales to hundreds and thousands of processors. We also present an energy group ''batching'' algorithm that greatly enhances cache performance. Our conclusion, after considering cache performance, storage limitations and dependencies inherent in the physics, is that a model that uses a combination of message-passing and threading is superior to one that uses message-passing alone. We present numerical evidence to support our conclusion.
- Research Organization:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- USDOE Office of Defense Programs (DP) (US)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 14136
- Report Number(s):
- UCRL-JC-133787; DP0101031; DP0101031; TRN: US0110966
- Resource Relation:
- Conference: Mathematics and Computation, Reactor Physics and Environmental Analysis in Nuclear Applications, Madrid (ES), 09/27/1999--09/30/1999; Other Information: PBD: 12 Apr 1999
- Country of Publication:
- United States
- Language:
- English
Similar Records
Experience with mixed MPI/threaded programming models
Data Locality Enhancement of Dynamic Simulations for Exascale Computing (Final Report)