Ask HN: Is it a good time for microkernel *nix OS

[smeg_it]

I forget exactly, but I believe that the original reasons the macrokernel was chosen over the microkernel are largely moot now. The microkernel was always more secure, if my memory serves. That is more important now than it ever was. I need to check but aren't we still behind on full 64 bit adoption? I would be think that we be better achieved through a newer kernel/OS. Lastly, AIs are still relatively new and not as locked down and/or expensive as they may become for specific purposes. Couldn't they help reverse engineer drivers? Which I would think would be one of the largest obstacles. I've read about Redox in the past, but it seemed to be a one man project that was largely abandoned for years and after 10 years it's still not out of beta, if I'm not mistaken.

I'm a middle aged +, amateur user/enthusiast. So, I'm really asking. Lastly, I wonder if the Linux kernel is like the Weasley house in Harry Potter i.e. kind of a monster with parts on top of parts , with the whole being more unruly and although functional is a bit of an unruly mess at this point that doesn't resemble anything like a good overall design like a good "new" house could be. Again, just asking. Most of this is well beyond my understanding.

[dc396]

Maybe it could be called Mach (https://www.cs.cmu.edu/afs/cs/project/mach/public/www/mach.h...) :)

More seriously, I believe the historical challenge with microkernels for Unix(-like) operating systems was the performance hit taken when transitioning from the user processes to the microkernel.

[smeg_it]

First thanks for the response.

"performance hit taken when transitioning from the user processes to the microkernel" is this still a large issue from the point of view of a desktop/laptop i.e. less embedded or other low power systems?

[dc396]

I haven't really been following microkernels for a (very) long time so it's entirely possible that the performance concerns have been overwhelmed by Moore's Law. IIRC, the problem was that microkernels had to do WAY more syscall-equivalents (regardless of how they're done) than monolithic kernels and the constant context switching slowed everything down, but if everything is orders of magnitude faster, maybe it isn't as much of an issue.

[mech422]

or SeL4 https://sel4.systems/ or Sculpt https://genode.org/download/sculpt

[smeg_it]

I had to look and skim to re-familiarize myself with SeL4. It's and older project that really never made it, from just a quick glance. It seems to basis or at least inspirational to more modern kernels. From the wiki, that seems to include Redox. I couldn't find much about sculpt, but it was submitted to distrowatch in 2018 and is still under the "not ready" section of their waiting list. I don't think there is really any usable FOSS OS using a microkernel. Last I remember reading, the ones most in use are highly specialized commercial projects. As far as I know Redox was the closest but it really hasn't been really actively developed (again, correct me if wrong). It hasn't been abandoned i.e. there seems to be development, but it's a long way from an end user OS at this point, from what I remember reading and no big strides have been made (again, maybe wrong)

[LargoLasskhyfv]

You know? Depending on the hardware you have, you could download a bootable image of Genode, put it on a USB-keychain, and 'sculpt' it from there, in 'live' mode, without touching whatever you have installed on that system. Some, or even most people developing it using it as their daily driver. By means of a ported Virtual Box, and running anything mot covered under that.

I'm not using it more, because I didn't want to touch my installation, because it flies like a mad bat out of hell :-)

But having a look, clicking around, while having the documentation available, trying things, enjoying it, can be done easily by live-booting it from USB.

[smeg_it]

I'm sorry for the late reply, but I wanted to thank you for the reply. I did a quick scan of genode on wikipedia. I've heard of it and might have seen articles over the years, but it says it's a framework and it kinda says it was built for study. You kind of hit on my point. Microkernels have been around for many years, but there has never really been enough resources put in to make them viable for general desktop use, except for Minx but that was proprietary. From what I read, it wasn't made open source until too late.

[LargoLasskhyfv]

No need to be sorry. I'm not in a hurry :) I still think you've got the wrong impression by quickly scanning wiki.

What I meant to say was to make up your own mind by going here:

https://genode.org/ reading that stuff, maybe proceeding further to

https://genode.org/download/ and choosing one of the Pre-built Sculpt OS images from there.

Either to boot them in 'live-mode' on bare metal without touching your installation, or put them in QEMU to test-drive them there.

Maybe skim https://genodians.org/ too, or if you've got the time watching others doing that (instead of yourself) there in reverse chronological order:

https://www.youtube.com/results?search_query=genode+os&sp=CA...

Don't read (outdated stuff). DO!

[mech422]

Hmm - don't remember if it was micro-kernel, but AtheOS (1) was a REALLY nice OS developed by a single very talented developer. It included display server, networking, partial POSIX compliance and lots of goodies like GCC, Bash, etc. Haiku moved to using the Linux Kernel if IIRC and AtheOS itself seems to have vanished...

1: https://atheos.metaproject.frl/

[mech422]

from what I've heard - sculpt is pretty close to 'daily driver' territory, and genode is used in the companies commercial offerings so it gets continuous work. Its actually the next OS I want to play with ...

[mech422]

not sure if you've seen this... https://www.microkernel.info/ but it might be helpful :-D

[smeg_it]

I appreciate all your replies. I really do. I can't say I've read everything but I have looked, scan, read much of what you posted, in one way or another e.g. I usually start with Wikipedia.

I nothing so far has convinced me there are enough resources put into any project to make anyone of them a "daily driver" for a desktop/laptop or that there will be anytime soon.

I can't complain as it's more a wish and/or a hope. I really don't have the time, brains, or money to significantly contribute. It's just that I was trying to ask those that might have one or more of those, if it might be a good time for the "foss" community to seriously consider it, or if not why. So, I can understand better.