[TUHS] OS for IBM PC (was: Algol68 vs. C at Bell Labs)

Larry McVoy lm at mcvoy.com
Tue Jul 5 04:13:30 AEST 2016

QNX, which wasn't Unix compat at the time but sorta close, in the mid
1980's was very very small and ran just fine on a 80286.  If my memory
serves me correctly I had 4-10 people logged into that box on terminals.

QNX, at least until they put real posix conformance in it, was a really
tiny micro kernel with the rest of the os in processes.  It fit in a 
4K instruction cache with room to spare.

QNX, in my opinion, is the only really interesting and commercially
proven microkernel.

On Mon, Jul 04, 2016 at 12:54:15PM -0400, Norman Wilson wrote:
> Greg Lehey:
>   And why?  Yes, the 8088 was a reasonably fast processor, so fast that
>   they could slow it down a little so that they could use the same
>   crystal to create the clock both for the CPU and the USART.  But the
>   base system had only 16 kB memory, only a little more than half the
>   size of the 6th Edition kernel.  Even without the issue of disks
>   (which could potentially have been worked around) it really wasn't big
>   enough for a multiprogramming OS.
> =====
> Those who remember the earliest UNIX (even if few of us have
> used it) might disagree with that.  Neither the PDP-7 nor the
> PDP-11/20 on which UNIX was born had memory management: a
> context switch was a swap.  That would have been pretty slow
> on floppies, so perhaps it wouldn't have been saleable, but
> it was certainly possible.
> In fact Heinz Lycklama revived the idea in the V6 era to
> create LSX, a UNIX for the early LSI-11 which had no
> memory management and a single ca. 300kiB floppy drive.
> It had more memory than the 8088 system, though: 20kiW,
> i.e. 40kiB.  Even so, Lycklama did quite a bit of work to
> squeeze the kernel down, reduce the number of processes
> and context switches, and so on.
> Here's a link to one of his papers on the system:
> https://www.computer.org/csdl/proceedings/afips/1977/5085/00/50850237.pdf
> I suspect it would have been possible to make a XENIX
> that would have worked on that hardware.  Whether it
> would have worked well enough to sell is another question.
> Norman Wilson
> Toronto ON

Larry McVoy            	     lm at mcvoy.com             http://www.mcvoy.com/lm 

More information about the TUHS mailing list