SOFA launcher

The tool sofa-launcher ease the scripting of numerous SOFA simulations. This can be done from XML or python scripts. To accelerate the processing of the simulations the script has the ability to run the simulation either: sequentially, in parallel or on a cluster.

There is two options to use it depending on your needs:

• You want to run a lot of simulation from you own python script. In this case you should look at the file “integration_example.py”. This example shows how to start simulation in sequence, in parallel or using a cluster. The example also shows how all the simulation results are returned back to your script so you can implement your own number-crunching processing on them, e.g.: plotting graphs with matplotlib.

• You don’t want to write your own python script but still want to start a lot a lot of simulations you should have a look at the sofa-launcher.py application. Example:

./sofa-launcher example.json

This application is controlled via a json configuration file

{
  "files": [
    "example.scn",
    "example.py"
  ],
  "variables": [
    {
      "GRAVITYXML": "1 2 3",
      "nbIterations": 1000
    },
    {
      "GRAVITYXML": "2 3 4",
      "nbIterations": 1000
    }
  ],
  "launcher": "parallel",
  "numjobs": 5,
  "resultsstyle": "results.thtml",
  "resultsfile": "results.html"
}

with:

• files: The sofa files with your scene
• variables: Some values in your scene can be changed automatically in each run of your scene
  You can use any name as $MYVALUE…it depend on what you put in your scene.
  in my example scene I replace the $GRAVITYXML with a different value at each run
• launcher: Indicate to launch the simulations in parallel (other choice are sequential or ssh)
• numjobs: with a maximum of 5 simulation in parallel
• resultsstyle: Name of the report file template, here html but it could be anything else (eg latex array)
• resultsfile: The file generated from the template and the results

NB:

• To run the distributed version of the launcher you need all hosts to share directories as well as being able to login using ssh+key (no password login).
• Depending on the files you are using in your scene you may need to protect certain symbols: see https://pythonhosted.org/Cheetah/users_guide/