Provide tools for parallel processing

[RemiLehe]

There are several steps.

Parallel processing on one node with openPMD-api

Provide example/instructions on how to process fields/particles in the WarpX documentation.

• For fields: provide example on how to load the fields, coarsen them and plot them. Or also: compute total field energy (proxy for laser energy)
• For particles: manual filtering and histogramming

• Provide example on which dependencies to install

(We should decide exactly where in the documentation this should go.)

Parallel processing on several nodes with openPMD-api

Provide example/instructions on how to:

• allocate a Dask cluster (e.g. through JupyterHub)
• make sure that Dask is using the cluster

Parallel processing on several nodes with openPMD-viewer

Provide an interface to use a Dask cluster in openPMD-viewer:

ts = OpenPMDTimeSeries( path_to_data, dask_cluster )

Parallel processing with GPUs

Provide instructions of all of the above worflows, but with RAPIDS objects, that run the DASK workflows on GPU (e.g. using cuDF)

Other tools (e.g. multiprocessing)

[RemiLehe]

Note that NERSC's JupyterHub has a Dask kernel, which I think is meant for this type of interactive, parallel work:

[ax3l]

That is a good point to document.

We need to close the gap of what the infrastructure provides to what we want to run:

• https://openpmd-api.readthedocs.io/en/0.16.1/analysis/dask.html
• https://openpmd-api.readthedocs.io/en/0.16.1/analysis/rapids.html

I think we need something like an overhaul of the Perlmutter Jupyter environment (also, let's add a script to set it up) and likely a small mini-notebook that starts up Dask nodes.