Modeling the Linux page cache for accurate simulation of data-intensive applications