A strategy for running large scale applications based on a model that optimizes the checkpoint interval for restart dumps