I think that it is exciting that people are still working on Hurd and there is a new release synced with Buster. I think there is latent potential in microkernels and that we may actually be at in position to exploit some of the benefits such as heightened security and stability. While at the same time the performance cost of context switching could be reduced due to the higher number of cores on modern chips (especially compared to the hardware available in the 90s). One of the biggest adoption challenge any new operating presents is hardware compatibility and well a reason to try it. 80% software compatibility with Debian is a big accomplishment, so this is probably worth spinning up a virtual machine to play around with.
Well I've been hearing about those benefits for at least 20 years. None yet actually sighted in the wild. Any attempted explanations from the microkernel religious zealots has been thoroughly unconvincing (because mach is slow, so?). Maybe you're right, I'd love to see it if you are. But while prior performance is no guarantee of the future, as they say in the funds management adverts, a pattern of repeated failure is something maybe not to completely ignore. Is there a reason for it that is being overlooked or glossed over. (In mutual funds it was and is M.E.R.) Stallman talks about hurd being really hard to debug, is that a thing? Or is there some kind of combinational explosion using multiple servers a message passing that isn't there with a monothlithic kernel? Something else entirely?
Anyway I'll go back to the american monolithic kernel conspiracy to destroy OS research and keep the Europeans out and ask the brothers if they can think of anything. (That's a joke, right? Yet I've heard it said in the absence of irony...)
Well Intel chose minix to run their often despised management engine, so their engineers saw some benefit there to the microkernel architecture. Google is putting at least nomimal developer resources into Fuschia. And the L4 microkernel and derivatives displayed that microkernels aren't doomed to be slow.
I think that Linux is probably going to be the dominant free software kernel for quite some time especially since it has finally gotten to the point of at least receiving nominal driver support by hardware manufacturers. So Hurd will be a curiosity for now but considering its history it is still very cool that development continues. Whether it will become useful in ways that Linux is not remains to be seen.
And General Dynamics[1] bought OK Labs[2] in 2012, so presumably OKL4 or a descendant is in use there, eh? (In addition to "over 2 billion mobile phones".)
Looks interesting. My former colleagues worked on porting osx to l4 more than a decade ago.
That still doesn't look like a microkernel based, multi server os to me and does not claim to exhibit those of touted advantages. This reminds me more of dresdens live demo cd from about 2006. Great stuff, but more of a virtualisation layer than an os.
Just out of curiosity, are you saying you don’t think the OS running on the Secure Enclave processor (which is separate from the main OS) looks like a multi server microkernel, or are you still referring to the osx on L4 Experiment you mentioned?
Show me those benefits on the wild. Intel's use of Minix, Qualcomm's use of l4 aren't showing me those benefits yet. If that's changed and they have I really do want to see it. Pointing out i haven't yet shouldn't be a sin, it should be quick to prove me wrong with some reasonable links with production ready more secure, more robust OSes. I'd pay performance for security and reliability in many instances. But AFAIK I can't.
You may also recall hearing about the benefits of GC for the 40 or so years before it became mainstream (because GC is slow, right?).
When advantages exist, they will eventually be exploited. It may take new research (better algorithms), new hardware (faster processors), or a new context (internet security), and those take time, but the original reasons for inventing the technology don't expire. As far as application of technology goes, a couple decades is not very long at all.
Nearly every computing technology I use today was loudly rejected by the mainstream, right up until it wasn't. Being unpopular seems to have no impact on the eventual success of computing technology, if it's a good idea.
When you have to resort to name-calling ("religious zealots") to explain why you won't look at actual advantages in computing, it makes me even more convinced that it's the correct approach, and will eventually win out.
my applogies. I felt i responded with precisely the same tone as the comment's parent. Pointing out a lack of supporting evidence seems reasonable to me, but ymmv.
These are small os's, they're great and useful. There are others. My former, now banrkupt, employers claim L4 code I contributed to is running on over a billion phones. is uTron, eCos etc still dominant in that space?
This is not showing the touted benefits of microkernel based, multiserver operating systems. This is not a secure operating system, this is not a robust operating system, this is not self-healing drivers. Is very small, monolithic a better description of what these are?
Do something like spin up a web-server on a multi server OS and demonstrate it's very hard to hack compared to apache on linux. Make it convincing enough so people use it. DJB did this with qmail and it really is convincing. How long have mircorkernel proponents been talking up improved security, robustness and resilience? 25 years? More? So we've had time and a mountain of engineering resources, including some of my own. It's reasonable to ask to see it. If we can't, it's reasonable to ask why not yet. It really is reasonable to ask.
In my opinion it's a sign of weakness to simply belittle anyone who does. But hey, this is a weekend microkernel HN story perhaps mostly read by the microkernel faithful so reasonable may not be the way forward, easier to just to act as though it's somehow completely unreasonable. Because anyone who hitches their wagon to the next big thing desperately wants it to succeed, and why not? Promotion is important for mindshare and so on. I have sympathy but I think Feynmann summed it up correctly:
"For a successful technology, reality must take precedence over public relations, for Nature cannot be fooled."
I'll say it again so the point isn't lost:
I would love to see those benefits in the wild. I really, really would.
You repeatedly crossed into snark in this thread, and fed the flamewar. Please don't. It's against the site guidelines, and we're trying for something better than that here.
Again beautifully condescending but that isn't really terribly convincing. Good to see you go right ahead with personal assumptions rather than engage with content because this tells me you are unable to do so. This kind of zero engagement while going straight to various kinds of subtle or less so ad-homonym attacks annoy me. They are contemptuous in all aspects of life.
What is your precise claim here? I state there isn't one. But you did use the word "hop" so there's that.
Point me at the microkernel based OS, using any microkernel, l4 or any other, that exhibits the standard set of microkernel benefit claims around security, robustness, self-healing drivers and has launched a thousand academic papers and all the rest.
Please. Really. I would like to see it. Even if it doesn't exist yet I'd like to see it soon. Really.
Condescension is a piss-weak technique exhibited by those who know they don't have a viable argument. Reality is that which continues whether you believe it or not. I say nothing about what seems to me about you from what you've said here because that's just plain rude.
I'd love it if we got better operating systems by any technique. I'd be very happy if came via microkernels. But it hasn't and that is just reality. Again, show otherwise if you can. I've stopped following it, maybe it exists?
But good on you for going straight for the personal attack and making broad, unspecific claims while offering no substantiation. "industry will keep on keeping on. Good things are to be favoured."
DriverKit on OSX. I thought the days were gone when people claimed osx is a microkernel based OS because it has mach code in it. https://lwn.net/Articles/703785/ doesn't make linux a microkernel. Nor dpdk, openonload nor lwip. I don't recommend hopping.
At least 25 years and these are the straws we clutch.
Maybe ditch the bad attitude, the belittlement, the condescension and try to understand why we are here? What is the actual reason that we are here? Because in understanding that it might be solved and that might be useful. Pretending we're somewhere else and anyone who says otherwise is a heretic who must be silenced is a sure way of making no progress.
You started a flamewar and then perpetuated it. We ban accounts that do that. Would you mind reviewing the site guidelines and not doing that on HN? We're trying for better than that here. Also, we don't want this place to burn to a crisp, which is what eventually happens with flamewars.
"Be kind. Don't be snarky. Comments should get more thoughtful and substantive, not less, as a topic gets more divisive."
I reject the idea that I started a flame war. I felt I followed the guideline above and did it in the face of pjmlp being personal, dismissive and non-substantiative while pointing out that this was happening. I did my best but if you say I failed because he hit his mark and got a rise out of me then I accept that and apologise unreservedly. If you feel the need to ban me for this infraction then you must of course do so and I will be at peace with that decision. I am completely aware of making points with evidence that are extremely unfashionable in certain tech communities. HN at its best works well doing this as it has for me in the past. Not point scoring but actual discussion, substantiated with evidence where one can learn something. A million "+1" messages aren't worth reading after all. I'd say the same for "you're wrong" messages and "you've obviously got a personal issue" messages as we saw here.
I'm assuming pjmlp got a similar message from you, possibly a stronger one? It would seem odd to me otherwise but I suppose it isn't really of any consequence.
I've posted a comment asking the other user not to be snarky in comments here, but honestly your comments went much further over the line than theirs did.
By the way, it's clear that you know a lot and have a lot of experience with this topic. That's great! We want people like you commenting on what they know about. But it's necessary to keep yourself within the site guidelines while doing so—they're written the way they are from long experience with what makes internet conversations and communities workable. That doesn't make it easy. We all know how annoying it becomes when the person we're arguing with doesn't 'get it' and seems to stubbornly resist our attempts to reason with them. A lot of the time they're probably feeling the same way. The only solution is to restrain oneself, whether the other person is doing so or not.
personally, I think people who are interested in hurd should take a look at redox. Its a new microkernel with similar design goals that uses rust as its implementation language.
I am excited to see how redox evolves, but unless they start writing, virtualizing, or porting drivers it is not much more than an experiment in how well Rust can handle OS programming.
Hurd is part of the GNU system; it's the intended kernel for the GNU userspace. Linux has been used as an intermediary because Hurd isn't ready for widespread use. But the idea has always been that Hurd would replace Linux and create a GNU system that is entirely under the FSF.
Beyond the philosophical differences, another difference is that Hurd is a microkernel and Linux is monolithic. Hurd can be considered a research project for exploring microkernels. The most well-developed microkernels are not open source or free.
> But the idea has always been that Hurd would replace Linux and create a GNU system that is entirely under the FSF.
I don't think that's the case. The FSF has long been satisfied that Linux meets its goals for an OS kernel, and the Hurd project has changed somewhat from "this will be the final piece of the GNU operating system" to "this is something we're working on to explore microkernel design." I think its developers have given up on it ever replacing Linux, as it's very behind in hardware support, and the gap is only ever growing rather than shrinking. For example, it doesn't support multicore or 64-bit userland (in a time when projects are pulling 32-bit userland support!).
Please don't post dismissive comments like this. If you know more, share what you know so the rest of us can learn. If you don't want to do that or don't have time, that's fine, but then please just don't post anything.
The wankel was actually like Hurd at one point, everybody thought it would be the next big thing and many car manufacturers had wankel engines. Mazda stuck it out and funded it into production.
> There are currently no plan for 64-bit userland, but there are plans for 64-bit kernelland with 32-bit userland, which will notably permit to efficiently make use of more than 2 GiB memory and provide 4 GiB userland addressing space. Work on this is currently in the master-x86_64 and port-amd64 branches for GNU Mach.
> That being said, you can always run a 32-bit version on a 64-bit machine, it just works, processes are just limited to a couple GiB available memory.
I don't really understand how all this stuff works at the kernel level, but if you have a 64-bit kernel what would prevent you from also having 64-bit user programs?
There is quite a lot of surface area between a program and the kernel - C library, dynamic linker, system call interface, memory layout, droves of permission and sanity checking logic etc. that would need to be updated too. A 64 bit kernel is a first step
That's ironic, given that last I heard anything about Hurd, it was that partition size was constrained by bus width (i.e., it couldn't access disks larger than 2 GiB on most computers at the time), but since everyone was moving to 64-bit architectures soon enough, that wouldn't be an issue. (Not sure where I heard that particular excuse, but the explanation why is here: [1])
32-bit userland makes Hurd basically unusable for many server applications, where multi-gigabyte in-memory lookup tables are critical for performance.
Does the Hurd team go around lecturing people who call it "Hurd" about how it's actually "GNU"/Hurd, how Hurd is the kernel and that's only one part, or is that only when the kernel wasn't their idea?
Hurd is a GNU project so it doesn't even really need a second mention, it only needs that mention because there is a competing kernel (Linux) that is used with the system. Otherwise it's just known as the GNU system. The kernel is pretty irrelevant for most users, we don't call it Android/Linux or Windows/NT for a reason. A kernel is not an operating system. Calling the whole OS Linux is both inaccurate and does not introduce anyone to the whole reason why the GNU project was started, which is having the freedom to really own and control your computer. Linus Torvalds doesn't give a shit about that and is happy with chromebooks as long as they use his kernel. Freedom is irrelevant to him and most of the "open-source" community. That should be reason enough to not call the entire system Linux.
You're argument holds water, but the problem is when we start getting into OS's that actually don't ship any of the GNU userland. The classic example is Alpine Linux. Do we call that BusyBox/Linux? WSL 1 was a super weird anti-example where Microsoft skipped out on the Linux and shipped GNU stuff and still called it Linux. I think the colloquial naming convention became something like Kleenex or Google where we default to blowing our nose with (generic usage, not brand usage) "kleenex" with off-brand tissues or had our parents (generic usage, not brand usage) "google" something on Bing (as thats the default search engine of Windows).
Depends on what you mean by Linux. The ABI Microsoft were attempting to emulate was absolutely Linux. (This API business is exactly the issue with Oracle vs Google over Java).
I assume Linus is a pragmatist who doesn't want to torpedo his own product by putting a slash and abbreviation in its name and forcing the whole world to only call it that. I think accusing him of not caring about freedom because he balances pragmatism and ideology differently is throwing out the proverbial baby with the bath water.
I don't think it's nice to spit in the face of a person that hands you half of the cake you wanted. The GNU project has the freedom to get their own HURD kernel into a sufficiently working state and compete, which does not seem to be that trivial a task for all the years I've been following this drama.
I don't disagree on any particular point, but I'd love to be able to post about network stacks and system calls on the Internet without someone butting in with, "Actually... I think you mean GNU/Linux. Linux is just a kernel..."
For most users, so is the command line. So when a user is on a Gnome desktop, we can call it Gnome/Linux right? Since a normal user doesn't touch the terminal?
Hurd is the official kernel of GNU project, so essentially it's just GNU. You could call it Debian GNU but I guess since that's confusing they decided to call it Debian GNU/Hurd.
I mean "GNU/Linux" is supposed to mean "GNU with Linux" i.e. GNU userspace with Linux kernel. GNU/GNU makes some sense but clearly is redundant. It's just GNU as in GNU userspace and kernel.
Well apart from actually building and researching how to build an OS. This stems from a very very deep seated itch.
The answer is "because we can".
It's volunteer engineering; because they can! Also it is not possible to know how the fruits of innovation might materialise. There might be achievments coming from this in a serendipitous way. The Hurd is a process.
After the news about pip being maintained by 5 people, I wonder how many are working on hurd. (As a grown adult, I also envy and more than a little humbled by their ability to devote time to it.)
QNX is widespread in embedded systems, it's sadly closed source (it's a magnificent OS).
L4 is a family of operating systems, sharing an API, not code. The ancestor was built to prove that microkernels could be fast. SeL4 is a formally proved variant. They are open source. They are also used in embedded systems, like baseband processors.
Minix 3 is an academic project by Andrew Tanenbaum. A few eyebrows were raised when it turned out that it is present in all new Intel chipsets.
Symbian (you can still buy handsets), QNX, L4, INTEGRITY OS, GenodeOS, muen, Fuchsia.
Then you have the hybrid ones from Apple, Android with Treble (classical Linux drivers are referring to as legacy driver on the documentation) and Windows is kind of hybrid as well.
As of Catalina, Apple was very clear that the long term roadmap for their OSes is to move all drivers and kernel extensions into userspace, which will be a gradual process.
It's a natural evolution of any system - monoliths are ok only as long as you don't need much of reliability, security, productivity, cooperation, etc.
> monoliths are ok only as long as you don't need much of reliability, security, productivity, cooperation, etc
As a supporter of microkernels I feel this is more a list of side effects. I'd put it this way: monoliths are ok so long as any critical code stays very small and comprehensible in it's entirety by an individual. It's also respectful of things that are not microkernels.
Good microkernels designs operate on the same principle, to achieve reliability and security they keep their critical code very small. It's not invulnerable to bugs, but it is well understood that minimising this surface area is the first step in minimising bugs, the secondary effect is also focusing attention due to minimising total lines of critical code. Microkernels are an attempt to take this to an absolute minimum by adding a layer of abstraction that makes otherwise critical code non-critical.
As a mere enthusiast I feel like this is the most useful lesson to take away from microkernels in other software, not triple redundancy or fancy reincarnation servers, but the fact that scale breeds complexity breeds bugs. Making sure the critical parts remain lean and inspectable helps a great deal in all software even when the separation is not as strict.
HelenOS is an interesting open-source micro-kernel based operating system project primarily fueled by Charles University in Prague. It has a working GUI and USB support. It appears to be driven by various dissertations that add functionality see http://www.helenos.org/
Maybe if Hurd decide to rewrite their kernel in rust I would give a s@#!, but since it's still a none usable kernel after decades of debugging and development, maybe it's ripe for /dev/null ?
Can someone with more knowledge answer, is it possible to have one core in kernel mode and another core in user mode? If yes, is it quicker to message between cores or to context switch?
