🚀 go-pugleaf

In article <624@caslon.cs.arizona.edu> dave@cs.arizona.edu (Dave P. Schaumann) writes:
   >VM is NOT a solution.

   This is the second post that says VM doesn't do task protection.  I was
   taught that VM gives every task its own unique address space, and when a
   memory fetch was made, a combination of hardware/firmware mapped this into
   an actual address in memory (after a possible disk fetch).  I realize that
   any machine that needs a RAM: disk has no business running VM.  When I
   said VM, I was only thinking of the 'unique address space' aspect of it.
   I also realize that, as explained elsewhere, because of message passing,
   memory protection would break the current OS.  *sigh*.

There are two accepted meanings for the term "virtual memory". The
most common one (at least, the one I see most often) is "demand paged
virtual memory", which is a solution to the budget-constrained memory
problem. This does _not_ imply that every task has it's own private
address space. You can do demand paging on a large virtual address
space, and have all tasks live in that same address space.

The other meaning is the one you allude to, "mapped virtual memory",
where each task has it's own memory map in the MMU. No paging - or
even swapping - needed. OS/9 used to do this on the 6809, and Unix
does it on the PDP-11. Unix practically requires support for this
feature.

The two concepts are orthogonal, though most people who do demand
paging also provide mapping.

BTW, if you look around in the Unix groups, you'll find people arguing
that RAM: disks are wonderful things (even on ther demand paged
virtual memory systems), and trying to find drivers for them. And no,
I can't explain that either.

	<mike
--