[COFF] [TUHS] RIP Niklaus Wirth, RIP John Walker (fwd)

[COFF] [TUHS] RIP Niklaus Wirth, RIP John Walker (fwd)

[Dave Horsfall]

[-- Attachment #1: Type: text/plain, Size: 2738 bytes --]

Might interest the bods here too...

-- Dave

---------- Forwarded message ----------
From: Paul Ruizendaal
To: "tuhs@tuhs.org" <tuhs@tuhs.org>
Subject: [TUHS] RIP Niklaus Wirth, RIP John Walker


Earlier this year two well known computer scientists passed away.

On New Year’s Day it was Niklaus Wirth aged 90. A month later it was John Walker aged 75. Both have some indirect links to Unix.

For Wirth the link is that a few sources claim that Plan 9 and the Go language are in part influenced by the design ideas of Oberon, the language and the OS. Maybe others on this list know more about those influences.


For Walker, the link is via the company that he was running as a side-business before he got underway with AutoCAD: https://www.fourmilab.ch/documents/marinchip/

In that business he was selling a 16-bit system for the S-100 bus, based around the TI9900 CPU (which from a programmer perspective is quite similar to a PDP11). For that system he wrote a Unix-like operating system around 1978-1980, called NOS/MT. He had never worked with Unix, but had spelled the BSTJ issues about it. It was fully written in assembler.

The design was rather unique, maybe inspired by Heinz Lycklama’s “Satellite Processor” paper in BSTJ 57-6. It has a central microkernel that handles message exchange, process scheduling and memory management. Each system call is a message. However, the system call message is then passed on to a privileged “fat kernel” process that handles it. The idea was to provide multiprocessor and network transparency: the microkernel could decide to run processes on other boards in the same rack or on remote systems over a network. Also the kernel processes could be remote. Hence its name “Network Operating System / Multitasking” or “NOS/MT”.

The system calls are pretty similar to Unix. The file system is implemented very similar to Unix (with i-nodes etc.), with some notable differences (there are file locking primitives and formatting a disk is a system call). File handles are not shareable, so special treatment for stdin/out/err is hardcoded. Scheduling and memory management are totally different -- unsurprising as in both cases it reflects the underlying hardware.

Just as NOS/MT was getting into a usable state, John decided to pivot to packaged software including a precursor of what would become the AutoCAD package. What was there worked and saw some use in the UK and Denmark in the 1980’s -- there are emulators that can still run it, along with its small library of tools and applications. “NOS/MT was left in an arrested state” as John puts it. I guess it will remain one of those many “what if” things in computer history.