jqj@gvax.cs.cornell.edu (J Q Johnson) (09/19/86)
As those of you familiar with Peter Denning's article on VM know, he does not really address most of the issues we've been discussing, since he argues that one can make the addressing penalty for VM negligible (a claim that the supercomputer folks dispute). His article does, though, suggest an interesting excuse for segmented architectures: since in a pure segmented architecture the VM is more user-visible than in a typical paged one, it is a natural candidate for software architectures that provide more control to the programmer, while still retaining VM. Phrased differently, perhaps those who argue in favor of programmer- defined overlays should call them "segments" instead.