Models of parallel processing systems typically assume that one has ll workers and jobs are split into an equal number of k=lk=l tasks. Splitting jobs into k > lk>l smaller tasks, i.e. using 'tiny tasks', can yield performance ...