is there any good rule of thumb for determining how many tasks keep all execution units busy? or is it relatively trial and error on a per-machine basis (from some starting guess inferred from the machine specs)?

@cmith23, that seems reasonable to me! As we did on the first written assignment, we can estimate idle time for tasks based on our hardware/memory resources and the specific I/O operations we are running. From there, we can run scaling analysis and performance profiler tools to determine optimal workload structures. The links here and here provide much more details!