Even if you use madvise() for a large sequential read, the kernel will often restrict its behavior to something suboptimal with respect to performance on modern hardware.