Out-of-memory paging is the second line of defense. If the process memory is not satisfied, the operating system gets a little more aggressive. It now finds another process's memory page (which hopefully has not been used in a while) and writes or pages it out to disk onto the swap area (on Windows, that's the page file). This is known as a page out (po), or a physical page fault. Physical page faults are normal and acceptable up to a point. Your operating system administrator and experience on a specific platform will provide the best guidance.

Swapping is the final line of defense. Traditionally, swapping refers to all memory associated with a process being moved out of real memory and written to disk in the swap area. This is indeed unfortunate, because if the process needs to be run, all that swapped-out memory must be read-swapped-back into real memory. It's expensive from a CPU perspective and when really bad, also from an IO subsystem perspective.

On an Oracle database server, we don't want to see pages or processes being swapped out. You'll notice that when there is plenty of memory to go around, the swap-outs will almost always be zero.