Recommendation for algorithm selection

[Andreas Nicolai]

Suggested procedure when tuning simulation performance with MasterSim

1. Basic assumption: FMU coupling is not very stiff, e.g. no stability issues when selecting constant time steps and having sufficient performance, use Gauss-Seidel without iteration and fixed steps
2. When you see stability issues, first try to change the FMU evaluation order within the Gauss-Seidel method.
3. If that doesn't help, reduce communication time steps
4. If performance is too low and there is potential for adaptive time stepping (e.g. some time during simulation there is little happening), switch to iterative methods with adaptive time step methods, select a low number of iterations (max 5) and see if this improves performance
5. If you get a lot of convergence errors, try Newton method as a last resort.

Note: MasterSim cannot make use of FMU-Jacobian-Structure information yet and generating the dense system Jacobian requires many FMU slave evaluations.

Generally, monitor to the performance stats at the end of the simulation.

[Andreas Nicolai]

Also see articles on documentation page of project web page.