I was just thinking this morning, to write a blog post with a small tour of niche-but-functional operating systems. Tribblix and Dragonfly BSD were at the top of my list. I've been also thinking Haiku, RISC OS, 9front, MINIX... They've all been around for decades, and usually fly under the radar.
I really like (and have used in production) the ability to run Solaris Zones (containers) and "LX zones" which are Linux containers side by side on the same hardware.
Many installation problems show up in the handling of USB3 and related quirks; I have a lovely AMD 8350 system that I can't install from USB onto because of some XHCI problem. The USB stick will boot the kernel, but then the kernel can't properly handle the USB stick after it has booted and can't see the installation media. I suppose I should bite the bullet and burn a CD to boot from, which should fix it.
That shouldn't be a fundamental problem with current Tribblix - due to this sort of problem the installer got modified so that once the bootloader has pulled the ramdisk off the usb stick it doesn't need it again (although if it can't see it it won't be able to install any packages from it, falling back to dragging them across the network).
The Illumos' family is an interesting one. I wish it was easier to get it installed on modern hardware. Any of my attempts with distros like OpenIndiana, Tribblix and OmniOS didn't go further than the boot menu.
I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability. Maybe the Oxide folks know more of what would be involved in such an effort.
That's the wrong solution to the problem, as it means learning a second kernel and all of its stuff and making compatibility shims, whilst still facing the real problem that is not software at all.
The right solution is actually explained the headlined WWW site, where Peter Tribble points out (in the About page and in the Use guide) that the significant constraint is that xe does not own the actual physical hardware to develop against. It's the usual story with small projects: good donated hardware, and developer time (and workspace, and food, and water, and housing, and electricity supply (-:), needed.
I agree with you in principle, and I understand the existing constraint. But I don't see developer time scaling up without those operating systems reaching (again) critical mass, and then having upstream wanting to write drivers for even more operating systems.
Which is fine per se. But I wish people could buy relatively new hardware (say, last 3 years) and be able to run FreeBSD or Illumos with most things working.
Not entirely sure what you mean (do you mean Threadripper, which some of use?), but this is purely a work machine for me. I regularly build and test our software on my Helios box.
It isn't exactly the same as our production machines (as you can imagine we have shared environments for this), but we've written in-memory simulators for most of our hardware components so it's good enough for most use cases.
Oh my bad, I meant "workstation" as in, a full desktop with a working desktop environment like XFCE, audio support, GPU acceleration for AMD iGPUs, etc.
> I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability.
It might be easier to take them from NetBSD; it wouldn't introduce the GPL licensing issue, and courtesy of their rump kernel system they're actually kind of designed for it.
> I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability
This isn't very interesting. It means you're constrained to the Linux design decisions, and you're wasting time debugging mismatches and poor design decisions.
It also creates a licensing issue. Part of the reason Linux has so many drivers is because the GPL requires all of the kernel code be provided to users under the GPL as well. If you link to that in your OS/distribution now all of your code must be published in the same way.
> Can't an hypothetical cross platform driver be MIT licensed and be part of both?
Yes, but if the goal is to reuse existing code that already exists in Linux then there's no reason to expect it to be under a permissive license. Some of it actually already is if I recall correctly, but you shouldn't expect that.
> Although I'd be more than happy with other OSes/distributions defaulting to GPL.
Illumos isn't GPL and and in the legal views of many people can't be compatible with it; CDDL/GPL (in)compatibility is a very long running issue, usually in the other direction with CDDL ZFS drivers in the GPL Linux kernel. (IANAL; I'm not asserting specifically that it is or is not actually compatible, just that a lot of people consider it to be incompatible)
You would have to look at the feature flags that your pool uses and see whether they match the flags that illumos supports. OpenZFS's feature flags are a superset of illumos's, so you'd be able to use a pool created on illumos on Linux/FreeBSD if that's what you're asking. https://openzfs.github.io/openzfs-docs/Basic%20Concepts/Feat...
Illumos hasn't been Solaris for a long time. In particular, illumos uses pool version 5000 plus feature flags, while the proprietary version is still on version 28 or so without feature flags. OpenZFS is also much closer to the exact code in illumos, as far as I know there are some differences these days because they don't stay exactly in sync but they're close-ish.
ZFS on Linux has effectively taken over as the standard version of ZFS. It got renamed to OpenZFS a few years ago after the FreeBSD project chose to start using it for their ZFS support rather than upstream illumos ZFS due to how much more active the ZoL project was than upstream.
> lightweight window managers are preferred over heavy desktop environments, the primary desktop option is Xfce, and MATE and Enlightenment are also available, plus many others
I would expected CDE as a first class citizen and maybe OpenLook.
And it says that it it maily for 32bit SPARC and 32bit X86 and later that "Important: 32-bit hardware support now completely removed.".
There are over 30 desktop options (including quite a few of the older window managers, which is a bit of a blast from the past).
I do include CDE and Open Look (the window manager and toolkit, at least). In both cases the add-on tools that were present in Solaris (like the whole of the DeskSet suite) aren't available, because they were never released in source form.
I was just thinking this morning, to write a blog post with a small tour of niche-but-functional operating systems. Tribblix and Dragonfly BSD were at the top of my list. I've been also thinking Haiku, RISC OS, 9front, MINIX... They've all been around for decades, and usually fly under the radar.
I really like (and have used in production) the ability to run Solaris Zones (containers) and "LX zones" which are Linux containers side by side on the same hardware.
Many installation problems show up in the handling of USB3 and related quirks; I have a lovely AMD 8350 system that I can't install from USB onto because of some XHCI problem. The USB stick will boot the kernel, but then the kernel can't properly handle the USB stick after it has booted and can't see the installation media. I suppose I should bite the bullet and burn a CD to boot from, which should fix it.
That shouldn't be a fundamental problem with current Tribblix - due to this sort of problem the installer got modified so that once the bootloader has pulled the ramdisk off the usb stick it doesn't need it again (although if it can't see it it won't be able to install any packages from it, falling back to dragging them across the network).
Why not use a PxE server? Mostly simple to setup.
The Illumos' family is an interesting one. I wish it was easier to get it installed on modern hardware. Any of my attempts with distros like OpenIndiana, Tribblix and OmniOS didn't go further than the boot menu.
I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability. Maybe the Oxide folks know more of what would be involved in such an effort.
That's the wrong solution to the problem, as it means learning a second kernel and all of its stuff and making compatibility shims, whilst still facing the real problem that is not software at all.
The right solution is actually explained the headlined WWW site, where Peter Tribble points out (in the About page and in the Use guide) that the significant constraint is that xe does not own the actual physical hardware to develop against. It's the usual story with small projects: good donated hardware, and developer time (and workspace, and food, and water, and housing, and electricity supply (-:), needed.
I agree with you in principle, and I understand the existing constraint. But I don't see developer time scaling up without those operating systems reaching (again) critical mass, and then having upstream wanting to write drivers for even more operating systems.
Which is fine per se. But I wish people could buy relatively new hardware (say, last 3 years) and be able to run FreeBSD or Illumos with most things working.
I run our (Oxide's) Helios distribution on a headless Ryzen 7900x with an ASRock Rack B650D4U motherboard. I use an Intel X520 NIC for 10GbE.
In general I'd expect Zen 3 and 4 to work, and Zen 5 either right now or very soon.
See https://github.com/oxidecomputer/helios-engvm
Thanks for the reply, I appreciate the link! Do you know if people are using it for workstations?
Not entirely sure what you mean (do you mean Threadripper, which some of use?), but this is purely a work machine for me. I regularly build and test our software on my Helios box.
It isn't exactly the same as our production machines (as you can imagine we have shared environments for this), but we've written in-memory simulators for most of our hardware components so it's good enough for most use cases.
Oh my bad, I meant "workstation" as in, a full desktop with a working desktop environment like XFCE, audio support, GPU acceleration for AMD iGPUs, etc.
Thanks again for the informative reply.
Ah gotcha! I asked internally, and the initial answer is "no", but if someone says yes I'll let you know.
> I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability.
It might be easier to take them from NetBSD; it wouldn't introduce the GPL licensing issue, and courtesy of their rump kernel system they're actually kind of designed for it.
Or OpenBSD, for at least one of those two reasons. I know Haiku reuses OpenBSD driver code in some places, and I'm pretty sure 9front does, too.
> I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability
This isn't very interesting. It means you're constrained to the Linux design decisions, and you're wasting time debugging mismatches and poor design decisions.
It also creates a licensing issue. Part of the reason Linux has so many drivers is because the GPL requires all of the kernel code be provided to users under the GPL as well. If you link to that in your OS/distribution now all of your code must be published in the same way.
Can't an hypothetical cross platform driver be MIT licensed and be part of both?
Although I'd be more than happy with other OSes/distributions defaulting to GPL.
> Can't an hypothetical cross platform driver be MIT licensed and be part of both?
Yes, but if the goal is to reuse existing code that already exists in Linux then there's no reason to expect it to be under a permissive license. Some of it actually already is if I recall correctly, but you shouldn't expect that.
> Although I'd be more than happy with other OSes/distributions defaulting to GPL.
Illumos isn't GPL and and in the legal views of many people can't be compatible with it; CDDL/GPL (in)compatibility is a very long running issue, usually in the other direction with CDDL ZFS drivers in the GPL Linux kernel. (IANAL; I'm not asserting specifically that it is or is not actually compatible, just that a lot of people consider it to be incompatible)
Sometimes it feels like keeping old systems alive is harder than keeping an old car running.
One thing I am curious about is its version of ZFS. Would it be incompatible with OpenZFS from a Linux/FreeBSD system?
You would have to look at the feature flags that your pool uses and see whether they match the flags that illumos supports. OpenZFS's feature flags are a superset of illumos's, so you'd be able to use a pool created on illumos on Linux/FreeBSD if that's what you're asking. https://openzfs.github.io/openzfs-docs/Basic%20Concepts/Feat...
You can create ZFS pools that are more or less compatible with anything.
Details here:
- https://vermaden.wordpress.com/2022/03/25/zfs-compatibility/
Solaris was the reference implementation of ZFS. Are you wondering if you could migrate disk pools from BSD/Linux to Solaris ZFS?
Illumos hasn't been Solaris for a long time. In particular, illumos uses pool version 5000 plus feature flags, while the proprietary version is still on version 28 or so without feature flags. OpenZFS is also much closer to the exact code in illumos, as far as I know there are some differences these days because they don't stay exactly in sync but they're close-ish.
Zfs on linux is its own project with a lot of development to make it run well on Linux
That answers the Linux part of your question
ZFS on Linux has effectively taken over as the standard version of ZFS. It got renamed to OpenZFS a few years ago after the FreeBSD project chose to start using it for their ZFS support rather than upstream illumos ZFS due to how much more active the ZoL project was than upstream.
> lightweight window managers are preferred over heavy desktop environments, the primary desktop option is Xfce, and MATE and Enlightenment are also available, plus many others
I would expected CDE as a first class citizen and maybe OpenLook.
And it says that it it maily for 32bit SPARC and 32bit X86 and later that "Important: 32-bit hardware support now completely removed.".
There are over 30 desktop options (including quite a few of the older window managers, which is a bit of a blast from the past).
I do include CDE and Open Look (the window manager and toolkit, at least). In both cases the add-on tools that were present in Solaris (like the whole of the DeskSet suite) aren't available, because they were never released in source form.
They have an overlay for both:
http://www.tribblix.org/overlay-cde.html
http://www.tribblix.org/overlay-openlook.html
I've been looking for a linux distribution that has olwm available but have had little luck. The closest I can find is a theme for icewm.
me as well, olvwm was last time in Debian Jessie, now only on NetBSD as package
Looks like Debian removed olvwm/olwm because they were incompatible with 64-bit and unmaintained.
https://tracker.debian.org/pkg/xview https://tracker.debian.org/news/1000764/removed-32p14-282-fr... https://bugs.debian.org/911787
It's never going to be maintained...
But there is a 64-bit port (which I ought to bring in to Tribblix)
https://github.com/ggodd/xview-64bit
[dead]