The transformation process should be split into N concurrent processes, each one handling 1/N of the input dataset.