Adrian (a.delete@this.acm.org) on September 27, 2021 10:40 pm wrote:
> David Hess (davidwhess.delete@this.gmail.com) on September 27, 2021 11:35 am wrote:
> > blaine (myname.delete@this.acm.org) on September 26, 2021 11:19 pm wrote:
> > >
> > > As someone who was there: Morse was under severe constraints in design of the 8086.
> > > I would not read anything into segments but as an expedient under constraints.
> >
> > I will add as an experienced 8080 and Z80 programmer at the
> > time, the 8086 segmenting scheme was seen as a great
> > improvement over contemporary developing solutions. Its
> > limitations only became a problem later when single
> > data structures exceeded the size of the entire 64 kilobyte
> > address range of the previous generation of processors,
> > and the alternatives (Zilog, others), except for a single larger address range (68000), were worse.
> >
>
>
> The 1978 Intel 8086 segmenting scheme was a great improvement not only over the existing microprocesors with a
> 16-bit address space, but also over what was available for the most popular minicomputers, i.e. DEC PDP-11.
>
> DEC PDP-11 extended its 16-bit address space to a 22-bit address
> space (i.e. 4 MB) using a very primitive paging scheme.
>
> Paging in PDP-11 and similar minicomputers had nothing to do with paging in VAX, 80386 and similar
> CPUs, where it was used for dynamic relocation, memory protection and virtual memory.
>
>
> Paging in PDP-11 was primarily only a method for address space extension, which was far more inconvenient
> to use than the segments of 8086, which could start at any 16-byte aligned address in the memory and which
> could be loaded together with in-segment pointers with a single instruction (JMP FAR, LDS, LES).
>
> In PDP-11 you could access simultaneously the equivalent of two 64 kB segments, a data segment and a
> code segment, but whenever you needed to access data or code from another part of the memory switching
> the right pages was a much more tedious and convoluted process than using far pointers with 8086.
>
> Also, the fact that 8086 had 4 simultaneously accessible segments instead of only
> 2 reduced a lot the need of switching segments in comparison with PDP-11.
>
>


I want to add that because of the difficulties of using the PDP-11 paging, in simpler operating systems, i.e. UNIX, paging was not used at all in user programs.

UNIX processes were limited in size to what was accessible without switching pages, i.e. an up to 64 kB text segment and an up to 64 kB sum of the data, bss, stack and heap sizes.

The entire up to 4 MB PDP-11 memory was used only by running concurrently many UNIX processes.

On the other hand on Intel 8086 there was no problem e.g. to use a 512 kB array in your program, if you compiled it for the large or compact memory models.