Two considerations for a new time-sharing machine:

Paging:
	inefficiencies
		shuffling
		whole image must be in core
	poor communication between jobs
	isolated by non-standard operating system
We now have gross inefficiencies because the time-sharing system
and all user programs must be entirely in core.  This is well-known,
it is also well-known that without paging we are locked out of improved
communication between programs.  We are also stuck with a non-standard
operating system.

indexing relative to PC
At first appearance, this would seem to be unnecessary with a paging
system because if pages are small enough (say 512 words), we have a
form of self-relocation.  We contend that this is not equivalent.  First,
a brief summary, for the sake of completeness, of why indexing relative
to the program counter is useful.  A program can be made self-relocating
so that it can be moved around at will.  Local references are made
relative to the program counter and global references relative to the
stack, etc.  The capabilities of paging would include those of indexing
relative to the PC if:
	pages were small
	the address space were infinite.
This index mode provides the facility to manage a very large potential
address space by simple means.  It seems to be the simplest way to
achieve extended addressing.