MIT provides MULTICS source and documentation - test-drb@ccmp.vtda.org

12 Nov 2007

...
  > >> But what machines could run the that
today?
 > > 
 > > Anything you port it to. 
The segmentation architecture would be hard to fake.  And
the idea of segments is so deeply ingrained, I doubt you'd
want to try.  However, the x86 does seem to have the necessary
segmentation structure.  Unfortunately, it's only got 32-bit
addresses.  Even in the 60s, the Multics machines had 36.

...
   That's an
interesting question.  I can't think of that many machines
 with a hardware ring architecture, and it's not something that can be
 easily faked with more traditional architectures.  It would be a fairly
 invasive "port".  
 It pains me to say it, but you could just throw CPU horsepower and
 memory at the problem. 
Yes and no.  In principle, the only thing you'd have to ensure
is that you gave control to the supervisor on a ring crossing.
So you could make all pages not in the current ring inaccessible.
On a page fault, apply the ring brackets in software and change
the page tables as a result.  At the very least, you'd take a
performance hit.

Having said that, somewhere in the back of my head, I want to
say that the x86 has at least part of the mechanism already.
So a port to it would be less painful than to a lot of others,
except of course, for the smaller address space.

Of course, I could be totally off.  I'm certainly no Multics
expert, and my memory of the details of the x86 segmentation
are rusty at best.

BLS