You actually need enough physical memory for the page swap and interrupt handler routines. If you don't have enough memory for that, then you end up trying to swap in and out the routines that are doing the swapping, and bad things happen.
So, technically, the minimum amount of physical memory you would need would be the size of those routines and the vector section, plus whatever size you want to run code out of. So probably something like a couple times the size of an instruction.
Of course, that is all dependent upon the fact that the OS was linked with this in mind during the build process. Which I'm very sure none of them major OSes are pre-built to do this. Even with Linux you'd have to do some customization.
Slashdot Mirror
You actually need enough physical memory for the page swap and interrupt handler routines. If you don't have enough memory for that, then you end up trying to swap in and out the routines that are doing the swapping, and bad things happen. So, technically, the minimum amount of physical memory you would need would be the size of those routines and the vector section, plus whatever size you want to run code out of. So probably something like a couple times the size of an instruction. Of course, that is all dependent upon the fact that the OS was linked with this in mind during the build process. Which I'm very sure none of them major OSes are pre-built to do this. Even with Linux you'd have to do some customization.