Parallelisation

ANUGA supports two parallel approaches that require minimal changes to a standard sequential script:

• OpenMP — multi-threaded shared-memory parallelism; enable by setting the thread count before creating the domain.

• MPI (distribute) — multi-process distributed-memory parallelism; create the domain on rank 0 then call distribute() to partition it across ranks.

The two approaches can be combined.

• OpenMP Parallelisation
• MPI Distribute Domain-first parallel workflow (distribute)

For long parallel runs, see Checkpointing for how to save and restart simulations from periodic checkpoints.

To diagnose load imbalance caused by wet/dry triangle distributions, see Load balance monitoring.

See also

Advanced Parallel Methods

GPU offloading, mesh-first distribution, offline partitioning, and uniform mesh refinement — for large HPC production runs.

ANUGA User Manual — Chapter 12: Parallel Simulation covers the MPI domain decomposition in depth, including ghost cell communication, scalability benchmarks, and HPC cluster setup.