Implement MPI load balancing #70
Description
Right now (k,w) jobs are separated between ranks in contiguous chunks. This can lead, for large and intensive jobs, for significant under-utilization of the resources and slow-down, as some ranks struggle with difficult computations while the others sit idle.
The simplest solution would be to randomize job assignment from the master (k,w) array -- in principle this should solve most of the load balancing.
The ideal solution would be an MPI scheduler system where jobs are allocated dynamically to free ranks by a master process.