This section describes an alternative way to run parameter sweeps on a separate network (i.e. an external cluster). If your full license is installed on one network (i.e. your office) and your extra engine licenses are installed on an external cluster, then the Job Manager feature of FDTD can't be used. Instead, you will use the full license on your local computer to generate a set of simulation files. These files are then transferred to the cluster. Very often, these jobs will be submitted to the clusters job scheduler. See the Running simulations section for more information on running simulations from the command line (i.e without using the Lumerical Job Manager). When the simulations are finished, the files are transferred back to the local desktop computer. You can then reload the data files and extract the requested information.
- Adjust model parameters and save a set of fsp files (a fsp file for each parameter value).
- Run all the simulations
- Load the simulation files and extract the required information
Advantages: The individual simulations can be sent to an external cluster or other workstations
Disadvantages: It is not quite as simple to run all the simulations
Step 1: Adjust model parameters and save a new fsp file for each parameter value
After setting up a parameter sweep as described at Parameter sweeps, rather than running the parameter sweep directly, simply choose Save to files, as shown below.
This will create a directory called with the same name as the original fsp file, in this case, usr_optimization_example. Inside this directory you will find 10 fsp file, called "test_sweep_1.fsp" to "test_sweep_10.fsp". Each file corresponds to a particular value of the parameters and there are 10 files because we chose 10 different values for our sweep parameter.
Step 2: Run the simulations
The simulations can be run on a cluster or extra workstations, without requiring a graphical user interface license. A shell or job scheduler is the most convenient way to run these simulations. For further details, see Resource configuration in the Installation category. In particular, Running from the command line using MPI for your operating system, or Job scheduler submission scripts (SGE, Slurm, Torque).
Step 3: Load the results
Once all the simulations have been completed, you can load all the results by choosing Load from files, as shown below.
This will reload all the results from the fsp files, as if they had been run on the local machine by pressing the Run button. You can then proceed with the analysis of the results as shown in Parameter sweeps.
|
Note: Re-running the parameter sweep Once the sweep is run, a new set of fsp files are saved to a folder. If the sweep is run again, then a set of fsp files will be saved to overwrite the previous files. To avoid overwriting the fsp files, use save as to change file name or current working directory before re-running the sweeps. |