# DANTSYS/MPI: a system for 3-D deterministic transport on parallel architectures

## Abstract

Since 1994, we have been using a data parallel form of our deterministic transport code DANTSYS to perform time-independent fixed source and eigenvalue calculations on the CM-200`s at Los Alamos National Laboratory (LANL). Parallelization of the transport sweep is obtained by using a 2-D spatial decomposition which retains the ability to invert the source iteration equation in a single iteration (i.e., the diagonal plane sweep). We have now implemented a message passing version of DANTSYS, referred to as DANTSYS/MPI, on the Cray T3D installed at Los Alamos in 1995. By taking advantage of the SPMD (Single Program, Multiple Data) architecture of the Cray T3D, as well as its low latency communications network, we have managed to achieve grind times (time to solve a single cell in phase space) of less than 10 nanoseconds on the 512 PE (Processing Element) T3D, as opposed to typical grind times of 150-200 nanoseconds on a 2048 PE CM-200, or 300-400 nanoseconds on a single PE of a Cray Y-MP. In addition, we have also parallelized the Diffusion Synthetic Accelerator (DSA) equations which are used to accelerate the convergence of the transport equation. DANTSYS/MPI currently runs on traditional Cray PVP`s and the Cray T3D, andmore »

- Authors:

- Publication Date:

- Research Org.:
- Los Alamos National Lab., NM (United States)

- Sponsoring Org.:
- USDOE, Washington, DC (United States)

- OSTI Identifier:
- 459820

- Report Number(s):
- LA-UR-96-4099; CONF-961249-2

ON: DE97003131

- DOE Contract Number:
- W-7405-ENG-36

- Resource Type:
- Conference

- Resource Relation:
- Conference: OECD/NEA meeting, Paris (France), 2-3 Dec 1996; Other Information: PBD: [1996]

- Country of Publication:
- United States

- Language:
- English

- Subject:
- 66 PHYSICS; 99 MATHEMATICS, COMPUTERS, INFORMATION SCIENCE, MANAGEMENT, LAW, MISCELLANEOUS; NEUTRAL-PARTICLE TRANSPORT; D CODES; PARALLEL PROCESSING; DATA TRANSMISSION; CRAY COMPUTERS; SUPERCOMPUTERS; THREE-DIMENSIONAL CALCULATIONS; ALGORITHMS; ITERATIVE METHODS

### Citation Formats

```
Baker, R.S., and Alcouffe, R.E.
```*DANTSYS/MPI: a system for 3-D deterministic transport on parallel architectures*. United States: N. p., 1996.
Web.

```
Baker, R.S., & Alcouffe, R.E.
```*DANTSYS/MPI: a system for 3-D deterministic transport on parallel architectures*. United States.

```
Baker, R.S., and Alcouffe, R.E. Tue .
"DANTSYS/MPI: a system for 3-D deterministic transport on parallel architectures". United States. https://www.osti.gov/servlets/purl/459820.
```

```
@article{osti_459820,
```

title = {DANTSYS/MPI: a system for 3-D deterministic transport on parallel architectures},

author = {Baker, R.S. and Alcouffe, R.E.},

abstractNote = {Since 1994, we have been using a data parallel form of our deterministic transport code DANTSYS to perform time-independent fixed source and eigenvalue calculations on the CM-200`s at Los Alamos National Laboratory (LANL). Parallelization of the transport sweep is obtained by using a 2-D spatial decomposition which retains the ability to invert the source iteration equation in a single iteration (i.e., the diagonal plane sweep). We have now implemented a message passing version of DANTSYS, referred to as DANTSYS/MPI, on the Cray T3D installed at Los Alamos in 1995. By taking advantage of the SPMD (Single Program, Multiple Data) architecture of the Cray T3D, as well as its low latency communications network, we have managed to achieve grind times (time to solve a single cell in phase space) of less than 10 nanoseconds on the 512 PE (Processing Element) T3D, as opposed to typical grind times of 150-200 nanoseconds on a 2048 PE CM-200, or 300-400 nanoseconds on a single PE of a Cray Y-MP. In addition, we have also parallelized the Diffusion Synthetic Accelerator (DSA) equations which are used to accelerate the convergence of the transport equation. DANTSYS/MPI currently runs on traditional Cray PVP`s and the Cray T3D, and it`s computational kernel (Sweep3D) has been ported to and tested on an array of SGI SMP`s (Symmetric Memory Processors), a network of IBM 590 workstations, an IBM SP2, and the Intel TFLOPs machine at Sandia National Laboratory. This paper describes the implementation of DANTSYS/MPI on the Cray T3D, and presents a simple performance model which accurately predicts the grind time as a function of the number of PE`s and problem size, or scalability. This paper also describes the parallel implementation and performance of the elliptic solver used in DANTSYS/MPI for solving the synthetic acceleration equations.},

doi = {},

journal = {},

number = ,

volume = ,

place = {United States},

year = {1996},

month = {12}

}