14 Nov
2022
14 Nov
'22
3:14 p.m.
> To be honest, I've forgotten many (most) of
the details. But that sounds
> about right. As I remember it, it was like SunOS. The key point was that
> the kernel only had one view of the memory system period, no FS
> buffer cache etc...which was a departure from many of the traditional UNIX
> implementations. IIRC they did not support BSD's mmap -- but check the
It sounds like they could have supported mmap()
easily. I'd love to see
this kernel, it sounds to me like it was SunOS with nicely done SMP
support. The guy that said he'd never seen anything like it before or
since, just makes me want to see it more.
"One view of the memory, period." That describes Multics.
Doug
14 Nov
14 Nov
4:50 p.m.
New subject: Book: "You Are Not Expected to Understand This"
On Mon, Nov 14, 2022 at 9:16 AM Douglas McIlroy <
douglas.mcilroy(a)dartmouth.edu> wrote:
"One view of the memory, period." That
describes Multics.
Yep -- but not surprising. There were a bunch of folks at DG that had
worked on a single-level store system (Project Fountain-Head) that had
failed [some of that story is described in Kidder's book]. They had studied
Multics extensively, and that system supposedly had taken a great deal from
it. I never completely knew what happened other than something went south
and it end up getting canceled. I later got to know of a number of the
MV10000 folks (who later joined us in the HW team at Stellar). But I
believe a bit issue with Fountain Head was both new HW and SW, whereas the
MV10000 allowed Nova /Eclipse code to 'just work.' [Kidder makes a big
deal out of this in the book].
The lessons and relevance to TUHS I think are a couple of things:
1.) When they did DG-UX, the SW folks had previously developed a kernel
with that view, so it's not surprising.
2.) Cost of >>solution development<< is the real end-customer/deployer
driver (economics beats architecture).
Clem
ᐧ
722
days inactive
722
days old
1 comments
2 participants
participants (2)
-
Clem Cole -
Douglas McIlroy