Once you have split your problem up into parallel chunks, how do you know how successful this has been? Is there more room for further parallelisation and optimisation, or have you achieved what is likely possible? For a problem how much improvement in performance will you likely achieve when parallelising it?
