2 1 since it is performed at run time there is a performance trade off between the efficiency of the new data decomposition for a subsequent phase of an algorithm and the cost of redistributing data among processors