Parallel Simulation - Adding Simulations to Queue

General
1

Hello,

I am using PLECS 4.8.2 and am running multiple simulations in parallel to speed up the simulation (essentially I am working on an optimisation problem and need to iterate through multiple combinations of values).

I am wondering whether it is supported to add simulations to the current simulation queue in the callback function? For example, if I want to increment a value and then rerun the simulation, can this be done by appending this into the simulation structs list and will the command pick it up, or would a second batch of simulations need to be run?

If there are any better strategies to do this, then please let me know,

Thanks for your help,

Finn

2

Hi Finn,

You have to wait for the plecs('simulate',opts) command to return before modifying anything in the opts structure. If you need feedback from a run to determine something in later runs, then you’ll have to run a second batch.

Regards,
Bryan

3

Hello Bryan,

Thanks for your help - this is what I plan to do.

Is there a way I can stop the currently running batch of simulations within the callback function? This might be useful for what I am looking to do (essentially there is no point running the particular batch if one value goes out of tolerance).

Kind regards,

Finn

4

There isn’t a built-in mechanism to stop a simulation from starting once the parallel simulation batch is initiated.

However, one could cause the other simulation runs to terminate at their start or during a run.

One approach - if your condition is not met then in your callback function create a file, say failed.txt. To prevent a simulation from starting, in the Model Initialization Commands you could check if the file exists and raise an Octave error (e.g. error('Failed.txt esists')).

If you have particularly long simulation runs, then you could also check via a C-Script if the file exists and raise an error in the C-Script or use an Assertion in the model. I would recommend a relatively “slower” execution rate for your C-Script as file operations could slow down the simulation (e.g. check every 100msec of simulation time, but in the context of your model).

You’ll need appropriate error handling in your callbacks to make sure everything goes smoothly, but I believe that approach could save you time.

1 Like