Restart behavior

[ppraegla]

What is the intended restart behavior or way to do restarts?

I make use of a lot of restarts for my simulations, i.e., I divide my simulations into small chunks to reduce the runtime of a single simulation run due to wall time constraints on the cluster. For this, I create a new output name for each restarted simulation

./build/release/4C tests/input_files/particle_dem_2d_adhesion_VdWDMT_walldiscretcond.4C.yaml output/out
./build/release/4C tests/input_files/particle_dem_2d_adhesion_VdWDMT_walldiscretcond.4C.yaml output/out_rs400 --restart 400 --restartfrom output/out
./build/release/4C tests/input_files/particle_dem_2d_adhesion_VdWDMT_walldiscretcond.4C.yaml output/out_rs800 --restart 800 --restartfrom output/out_rs400
...

This gives me a clear naming of the output files (.control, .pvd, -vtk-files) and works quite well to do further restarts. Sometimes simulations crash and I want to rerun the same simulation. That's when the restart gets confusing. If I rerun

./build/release/4C tests/input_files/particle_dem_2d_adhesion_VdWDMT_walldiscretcond.4C.yaml output/out_rs400 --restart 400 --restartfrom output/out

out_rs400.control already exists and 4C automatically chooses a new output name out_rs400-1. This is unexpected because I chose another output name. So, a subsequent restart now needs to restartfrom out_rs400-1. In the case the simulation is rerun again, the output name is now out_rs400-2, which makes it difficult to know what the correct restart name is.

I can see the advantage of not having to think about specifying a new output name and letting 4C figure out the correct restart file. But this makes it difficult to know which files are actually relevant and which files are remnants of failed simulations.

Questions

• How do others do their restarts?
• Is anyone using the restarts without restartfrom?

[amgebauer]

Thank you for opening the discussion. I also struggled with the behavior that 4C magically appends a suffix to the output, which makes automatic restarts (which I also use a lot) sometimes difficult. I would favor a solution that is more predictable.

I also always specify restartfrom and a new output prefix.

[rjoussen]

I also do a lot of automatic restarts and it gets really messy when restarting from the same old simulation twice (because one of them failed). So I also opted for never using the automatic naming in automatic restarts. Most of the time, I want to put the restart in a different folder also, because otherwise there are two many files in the same directory imo.

I also never use the automatic naming for manual restarts, because I find it not very intuitive. I think if any, the automatic naming should append something more informative (like the restart step) instead of just an index.