[TUHS] Is there a good, even definitive, list of reimplementations of the Unix kernel? What would good cut-off criteria be?

[Larry McVoy via TUHS]

On Sun, Apr 19, 2026 at 04:02:05PM -0400, John R Levine via TUHS wrote:
> >OSX uses a microkernel, Mach. Thing is more and more code got moved from user space into the kernel to speed it up. First OSF1 moved more code into kernel space, then NeXT, then Apple. But the kernel definitely started out as the Mach micro kernel.
> 
> That's what almost always happens, start with a microkernel, find that all
> the context switches are killing performance so you move all the pieces into
> one place and you end up with yet another macrokernel.
> 
> QNX is probably the most successful microkernel system.  It's even POSIX
> compatible.

It was successful because only 3 people were allowed to commit code to the
microkernel itself.  They benchmarked every commit, down to counting cache
misses for important stuff like context switching and flagging commits 
that increased cache misses.  That's what Dan Hildebrandt, one of 3 people
in question, told me.

I miss Dan, we were friends coming at the OS problem from different points
of view but we always managed to have great conversations.  He could point
out the benefits of a microkernel but acknowledge the costs.  It was never
personal with him, he was trying to strike a balance and I think QNX showed
that they did.

I have never seen that sort of discipline in any other OS team, including
the ones I worked on.

And I ran multiple people editing and compiling files on the pre-POSIX 
QNX on a 286.  Pretty amazing.
-- 
---
Larry McVoy           Retired to fishing          http://www.mcvoy.com/lm/boat