Parallel - SCONE - ReactorMC

No Parallel Block in Input

SCONE does not have a parallel { mode = ... } input block. Parallelism is controlled at compile time (OpenMP flags) and runtime (environment variables). There is no MPI, hybrid mode, domain decomposition, or load balancing as input-deck features.

How SCONE Parallelizes

SCONE runs in parallel using OpenMP (shared memory). Each particle history is independent, so histories can be distributed across threads. Parallelism is controlled by:

Compile flags: Enable OpenMP when building SCONE (e.g., -fopenmp for gfortran).
Environment variables: Set OMP_NUM_THREADS before running.
Input settings: pop, active, and inactive affect total work and parallel efficiency.

Compiling with OpenMP

SCONE must be built with OpenMP support. Typical compiler flags:

bash

# Compile SCONE with OpenMP support (typical flags)
gfortran -fopenmp -O2 -o scone scone_main.f90 [other sources...]

# Or with Intel Fortran
ifort -qopenmp -O2 -o scone scone_main.f90 [other sources...]

Running with Multiple Threads

Set OMP_NUM_THREADS before launching the executable. The value should match the number of physical cores (or logical cores) you want to use.

bash

# Run with 4 OpenMP threads (Linux/macOS)
export OMP_NUM_THREADS=4
./build/scone.out input.inp

# Or inline for a single run
OMP_NUM_THREADS=4 ./build/scone.out input.inp

# Windows (PowerShell)
$env:OMP_NUM_THREADS = 4
.\build\scone.out input.inp

Input Settings That Affect Parallel Performance

The pop, active, and inactive parameters control how many particles and cycles are run. Larger values increase total work and can improve parallel scaling, but also increase runtime and memory.

plaintext

type eigenPhysicsPackage;
pop 100000; active 300; inactive 200;
XSdata ce; dataType ce;

collisionOperator { neutronCE { type neutronCEstd; } }
transportOperator { type transportOperatorDT; }
geometry { type geometryStd; boundary (0 0 0 0 0 0); graph { type shrunk; } surfaces { ... } cells { ... } universes { ... } }
nuclearData { handles { ce { type aceNeutronDatabase; aceLibrary /path; } } materials { ... } }
inactiveTally {}
activeTally { fissionRate { type collisionClerk; response (fission); fission { type macroResponse; MT -6; } } }

Key Parameters

pop: Total particles per cycle (e.g., 100000).
active: Number of active cycles for tally accumulation.
inactive: Number of inactive cycles for source convergence.

What SCONE Does Not Support

The following are not available in SCONE:

MPI (distributed memory across nodes)
Hybrid MPI+OpenMP mode
Domain decomposition
Load balancing as an input-deck feature
parallel { mode = ... } or any parallel input block
settings { ... } input block

Best Practices

Match OMP_NUM_THREADS to available cores.
Use enough pop and active cycles for statistically meaningful results.
Ensure sufficient inactive cycles for source convergence.
For large problems, consider memory per thread when scaling.

Next Steps

SCONE is designed for extensibility through Fortran source code modification. The Extensions section explains how to add new tally clerks, surface types, or physics packages.

Continue to Extensions

SCONE Guide

SCONE Verification Notice

Parallel Computing in SCONE