>Well, the right thing to implement is POSIX real-time scheduling.

Jason,

It might be the right thing to do from a standards perspective, and
may be worth doing anyway, but (wearing a research hat) the bottom
line I see is, it just doesnt work very well for applications like
real-time display of compressed video, or audio, or (I'd imagine)
software modems.

Even Jason Hsieh's (? cant check the Stanford database and he's now at
Sun Labs) stuff isnt completely satisfactory.  Let alone the perennial
`Oh look, my big compile paged out a page my [sw modem,mpeg player] needs!'

Yes, sure, the industry as a whole wants solutions for these problems,
I'm just not aware of POSIX RT being a particularly good one.