I've always liked the concept of Fortran co-arrays (even though I don't really program in Fortran itself). Julia's distributed arrays are conceptually similar but based on "simple" socket implementation and hence probably (don't shoot me if I' m wrong) lower performance than Fortran co-arrays which usually run on top of MPI or GASNet or something similar. I've just stumbled upon
http://www.opencoarrays.org/ (code repo:
https://bitbucket.org/sourceryinstitute/opencoarrays) which is an open source C lib for implementing fortran co-arrays. It's meant for fortran compiler writers but it occurred to me that this could probably also be used to implement distributed arrays in julia on top of MPI or GASnet. I've seen direct MPI wrappers but this would be nicer since at a higher conceptual level & allowing to switch underlying transports (MPI or GASnet or something else) transparantly + it would integrate nicely with the julia parallelism via distributed arrays that's already there. So if running on a cluster/computer without opencoarrays/MPI runtime, distributed arrays would use the socket based transport,otherwise they would use opencoarrays.