Ask HN: Is it a good time for microkernel *nix OS | Hacker News Reader

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

3 pointssmeg_it14 comments

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.

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.

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

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)

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.

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.

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!

I could be wrong, but I get the impression your younger than I am by a good bit. I'm an aging genx'r. My excitement/energy is lower than what it was. As I've aged, I feel like I don't have as much time, and I know I don't have the energy I used to. I wouldn't mind putting it in qemu, but if the drivers aren't there, bare metal is out for me.

On another note, I'm still trying to figure out the right balance in a home environment for direct hardware vs virtualization. I have taken classes, but they mostly talk about production environments. Many, don't even mention bare metal. I did read something on here recently, about minimal code and that was interesting. I've never been an expert, but I've enjoyed linux for quite a while. I wan't an early adopter but found it about 2008 and never went back. I never got used to snap/flatpacks/appimages but I know there is a place for them. My problem was that it made the old admin commands much less useful. 'mount' had too much mess. I know there is grep -v but you shouldn't have to do tons of that when trying to get an overview. top/htop are good but now there are tons of threads. So even when using tree view, it's harder to see. Not to mention, the ones I've tried, often lack features or features that don't work in the image form. I've studied lots of stuff, some at work, and some at home. I like to think I've forgotten more than some linux users ;P of course that doesn't apply to everyone. There have always been smarter people that know much much more. That's why I like to ask here. I really don't like videos as much; I prefer reading. Sometimes videos do help me but to me, they are much more time consuming, than reading and sometimes I watch half a vid before realizing it doesn't have what I was looking for. Right now, I need to study and use LLVMs a bit. I'm very late to that party. If nothing else, for searching the internet. Search engines just don't work well anymore. I have lots I want to do and work-on, if there was a large movement towards the microkernel, I definitely would want to move to that, but I just don't think it's there yet. Still, if time, I'll play with something. Redox, seemed to be full of great ideas, but I think it's stalled. When time, I'll try to look more at genode and I do appreciate the links. The trouble is, for me, I've read about many interesting projects that just never really come to fruition. I know resources is a huge problem. For years I've had the impression that unlike the olden days, where one programmer could change everything e.g. Linus, Stallman, and scores of others I don't know offhand. One person is no longer enough to change the world in this area, well with the exception of billionaires, like the guy behind Ubuntu. Even if your not a fan, it did change the landscape. Of course, the guy that did Redox, came really close, from my reading. If it was finished withing a few years, I think it would have. Maybe it still will. They list about a hundred developers. In contrast, I think the linux kernel has a couple thousand, but that's from a quick search. That's just the kernel, not a kernel, full OS with tools. Anyway, again, I appreciate it but for now, I need to work on some other stuff today :)

I know I didn't mention containers or VMs but again, not sure where the balance is, for me. I've thought about using Qube OS but I don't know where or if it's right for me. I tried IBMs VOID OS, and it was a pain for me. Again functionality and configuration. Not to be confused with Linux Void, which is really my favorite OS, but I'm not on it now because I have less time now than I did and it did still require more effort than Ubuntu or Mint. When I get time, I'll either go back to void, or do something else with my systems. Ubuntu is really kind of went to our way or the highway with the snap mess. Gentoo was also a great OS. I tend to really like Portage Systems, when they work. It can be a real pain, when they don't. Freebsd was good but never really desktop friendly. That might have changed but there where always issues with basic desktop programs when I used it, years ago e.g. LibreOffice and such. It seems to me, it was a case of devs not using it on the desktop. From the forum, the devs mainly used mac, and maybe freebsd on a server which it was perfect for, but the desktop ports seemed to be broken often.

Anyway, if your younger than I, am enjoy it and exploring. It is/was fun for me. I used to go to my states Linux Fest every year. I also used to get Linux Journal before it died. I miss that magazine ;)

Me fell into the chaos of this world in the summer of 1969 :-)

I can understand the lack/loss of enthusiasm for trying things out.

It's just that I'm not that much constrained by time, or other things stressing me out, even if I have the time.

Or not having enough hardware to try things out, without compromising my daily drivers.

I just need to watch out for balance between physical activity and morphing into a hunchbacked nerd from being online so much :)

Regarding Genode, and daily driving, I'm not doing that ATM, though it blew my mind.

I can imagine using it as something like a very lightweight Qubes by means of bundling applications into Unikernels like NetBSDs RUMP, Unikraft, Googles Distroless, or even most minimal Alpine. And 'sculpting' their capabilities from Genode.

But that's a hassle. But maybe not more so than how it is done now in general. And probably way more secure.

How it really performs fully loaded over time remains to be seen, and I haven't done that, yet.

Anyways, I just nagged you with that thing because I have the feeling people don't know this, while talking about exactly the things it could enable, and wouldn't it be nice to have something like that? :)

Or IF they know it, only from reading/talking about it, which is not the same as having seen it in action, and playing around :)

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

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.

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 ...

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/

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?

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.

Reply on news.ycombinator.com