Skip to content
Draft. This manual is new and still being checked against the software as it is verified. Some pages will change. If something here does not match what you see, the software is right — tell us and we will fix the page.

Run in parallel

Decompose the case across cores so the solver uses the whole machine.

The free tier solves on one core. Everything below can be set, and the case is written correctly, but the run is clamped to the licensed core count and the Problems tab says so. Check what your licence allows under Settings → Licence.

  1. Select Simulation control.
  2. Under Parallel, set Cores.
  3. Read the block below it. It reports cells per core and states the efficient range as 30–100 k; it turns amber outside a wider tolerance either side. Too few cells per core spends the run on communication rather than solving.
  4. Set Decomposition if the default does not suit: scotch, hierarchical, simple, kahip, multiLevel or manual.
  5. Run as normal.

The application runs decomposePar, then the solver under mpirun, then reconstructPar, so what you get back is an ordinary reconstructed case. You do not decompose or reconstruct by hand.

Set it from the cell count, not from the machine. Divide the cells by a figure in the efficient range and use that many cores; more than the mesh can feed makes the run slower, not faster. The read-out on the pane does this arithmetic for you.

  • The Runs pane’s status line shows the solver, the core count and the executor.
  • The Log tab shows the decomposition and then the parallel solver starting.
  • The Problems tab carries no core-count warning. If it does, the run is proceeding on fewer cores than you asked for.