If you have a simulation job that outputs a single cache file or multiple cache files per frame, then you can just send it like every other job that renders images.
For the rrServer it does not matter if the output file of an job is an image file or a cache file.
In Houdini, you could also add a small mantra render after the cache and submit it as a default render and have a few preview images.
Once machine only:
You should set "Max clients at a a time" to 1 and disable "Render Preview First".
All frames at once:
Some simulations require that the whole frame range is simulated in one continuous execution.
You have to enable Sequence Divide and set the Sequence Divide - Min to 0 for these simulations.
It will render on one machine only.
If a render aborts/crashes, it starts at the first frame again.
You sometimes need to simulate some stuff which does not output a cache file per frame OR a cache file for the whole sequence (=job).
In this case you have two choices:
Enable the option "Do not check for frames" for the job.
The server sends the job to clients, but each frame only once.
It does not care if your simulation has written the cache or not.
If you use your own script to do this simulation, you can just write a "fake" output file at the end of your simulation function (it has to be at least 1024 bytes).
So you have a proof that your simulation function was running until the end and the server finds "frame" files to count.
You can use the executable rrCreatePlaceholderFile to create a 1024byte file.
Note: If all frames have to be rendered in a row by one machine, please set sequence divide to 0.
If you set Sequence Divide Min to 0, then the sequence is not divided.
All frames are send at once to one client.