looked at the MPI protocol/standard for non-local execution today for actually very petty reasons and became interested in the MPI-IO interface (which is a stub on wikipedia, even though the stub says "HDF5 exists because of this"). this paper https://www.mcs.anl.gov/~thakur/dtype/ is going into the author's definition of non-local data and how they optimize it (this paper is more famous: https://www.mcs.anl.gov/~thakur/papers/romio-coll.pdf). i think it's possible the pants process execution implementation could be doing some of these techniques in the paper if i try to think abstract. this isn't likely to be useful for anything, i just think that things like futures and rust didn't exist a while ago.