I was thinking, should there be any difference in expectation of the page fault rate on CPU-bound vs I/O bound processes? At first I thought maybe we could, since CPU-bound processes would likely be using more memory accesses per time quantum, so I expect it would move from locality to locality faster. At the same time, the CPU-bound process is probably given a larger working set... but this doesn't affect the fault overhead as it hits a new locality IF this wasn't pre-paged in. Is there actually any real difference in the page fault rates or am I just musing about something nonexistent? And if there is, how would it impact a real-world OS like linux?