Not new, this paper is from 2012. There's also a link to a google code project which is a 404 now. On the other hand, the MPI library that they compare against is still alive here:
Oh, thanks for finding this because I was curious about checking out the code (that's why I found the broken link after all). I did a quick google search but wasn't able to come up with anything.
Google code is in archive mode, but the tgz is still there.
And this code has huge advantages over MPI, if you read the paper. It's statically proven deadlock free, and it optimizes bad protocol specs to use async.
It's almost on the level of ATS or Pony, just not memory-safe and type-safe, and structs, unions and objects cannot be passed to parallel functions without some custom serialization.
Mpi is pretty cool, I did a silly hack on Linux combined with mpi to update the command name programmatically with the status/percentage complete so you could monitor progress of all the jobs via top.
https://www.open-mpi.org