They touch on an important point: Lots of “legacy” RISC stuff was easy to support when you could purchase older workstations/servers for free or very little money. The retro computer craze and recyclers have driven prices of these “retro” upwards making it difficult to support platforms that aren’t commodity and/or have strong corporate backing (IBM’s POWER)
Legacy machines that were common are also too slow to use for anything meaningful now. Well, or we put it another way, their power consumption to computing power ratio is so poor by now that they are only good as museum pieces, or for some legacy applications that cannot be easily replaced.
I would be all for dropping SPARC support in Illumos, because I want Illumos to live on.
This is pretty much where we’re coming from. It’s disingenuous for us as a project to say we’re supporting SPARC while integrating changes that break the SPARC build faster than the handful of folks maintaining SPARC machines can keep up. At the other end it’s a serious drag on certain kinds of innovation and new work on the x86 side when we have to consider how we might later make the new things work on SPARC as well.
For a little more context on this, there’s a mailing list thread as well as the IPD linked here.
While I know illumos developers probably have enough on their plate keeping up with development on x86, I’m still wondering if there has been any talk of porting illumos on any new non-x86 architectures in the future? ARM hardware is becoming easier to get your hands on, and increasingly standardized and common in the server-space.
I installed OpenBSD on my old AMD K6-2, because nostalgia, and was going to use it for a few things… except it’s slower than a Raspberry Pi, so I dropped that. It just sits on a shelf now…
I tried running Debian on this machine and the latest version I managed to run was 7, which is really outdated. If I really needed to use that PC, then OpenBSD still worked as expected (I am sure NetBSD would probably have worked well too). I think anything would run super slow regardless, even Alpine Linux.
My take, as someone who does big-endian RISC as a dayjob…
The systems being dropped (basically 2000s SPARC) are kind of on the hinterland in between “vintage computing” and now obsolete enterprise equipment.
However, these systems are like $200 cheap on eBay if someone cares.
These kind of platforms tend to get supported because there’s a sufficient community (w/ time and skill) for it (i.e the 68k community willing the LLVM port into existence and maintainers rising up), or a commercial interest (i.e many mainstream projects support big-endian stuff because IBM wants to make sure it works on z). It seems illumos had neither (well, there’s the Tribbix fork, but is that illumos proper?).
Yes, Oracle still sells lots of SPARC servers, and they are still used in some weird embedded applications as well. There are even new chips still in development. Definitely going out of style though.
To add to the trouble, Illumos doesn’t have much support for post-Sun SPARC hardware anyway. The codebase branched off of OpenSolaris in 2010, so it’s mostly frozen at whatever was supported then. As far as I know, there hasn’t been much (any?) interest in trying to get it running on newer Oracle SPARC hardware. The commercial users of Illumos (e.g. Joyent, OmniTI, Nexenta) have used it only on x86, and SPARC hobbyists aren’t buying recent $100k Oracle servers.
Perhaps half their user base but evidently not half their developer base. If the user base ratio really is that high then more of those users need to step up to support SPARC at a developer level
It’s difficult to get an exact estimate, but I seriously expect the number of SPARC illumos users in 2021 can probably be counted on both hands. There’s only one seriously polished distribution for SPARC systems of which I’m aware, which is Tribblix. Over the last ten years periodically individuals will emerge that say they want to add SPARC support to some other distribution, but inevitably the effort goes quiet after a while.
The reality is that the SPARC systems we have a hope of supporting are not high performing by modern standards, so building a distribution worth of software takes a long time just to run the builds – and that’s assuming the software all builds already which it likely does not without patches.
Peter intends to manage a fork for SPARC support in Tribblix, so users of that distribution will likely be no worse off than they are today. I don’t expect this decision will impact more than ten or twenty people at most. Most all of the rest of the user base is using x86.
They touch on an important point: Lots of “legacy” RISC stuff was easy to support when you could purchase older workstations/servers for free or very little money. The retro computer craze and recyclers have driven prices of these “retro” upwards making it difficult to support platforms that aren’t commodity and/or have strong corporate backing (IBM’s POWER)
Legacy machines that were common are also too slow to use for anything meaningful now. Well, or we put it another way, their power consumption to computing power ratio is so poor by now that they are only good as museum pieces, or for some legacy applications that cannot be easily replaced.
I would be all for dropping SPARC support in Illumos, because I want Illumos to live on.
This is pretty much where we’re coming from. It’s disingenuous for us as a project to say we’re supporting SPARC while integrating changes that break the SPARC build faster than the handful of folks maintaining SPARC machines can keep up. At the other end it’s a serious drag on certain kinds of innovation and new work on the x86 side when we have to consider how we might later make the new things work on SPARC as well.
For a little more context on this, there’s a mailing list thread as well as the IPD linked here.
While I know illumos developers probably have enough on their plate keeping up with development on x86, I’m still wondering if there has been any talk of porting illumos on any new non-x86 architectures in the future? ARM hardware is becoming easier to get your hands on, and increasingly standardized and common in the server-space.
Port it to OpenPOWER. Side benefit: Larry Ellison gurgle of apoplexy audible in most western states.
I installed OpenBSD on my old AMD K6-2, because nostalgia, and was going to use it for a few things… except it’s slower than a Raspberry Pi, so I dropped that. It just sits on a shelf now…
Honestly for vintage computers I like to just run contemporary systems on them. No sense torturing a VAX running NetBSD when it barely runs it.
I tried running Debian on this machine and the latest version I managed to run was 7, which is really outdated. If I really needed to use that PC, then OpenBSD still worked as expected (I am sure NetBSD would probably have worked well too). I think anything would run super slow regardless, even Alpine Linux.
By contemporary with a K6-2, think Windows 95. Or Debian potato, if that’s how you roll.
Gotcha. It had 98 before I installed Debian (then OpenBSD). 98 seemed to run OK :)
Potato was my first Debian - on a K6-2 as it happens
I think it was Sarge or Etch in my case. The K6-2 was my “home lab” machine, ran cups for printing and a bunch of other things. Good times.
Fond memories of my long-departed K6-2. I used that for the majority of my undergrad.
My take, as someone who does big-endian RISC as a dayjob…
Tribblix is a distribution of illumos, not a fork.
https://illumos.org/docs/about/distro/
Do they even make SPARC machines anymore?
Yes, Oracle still sells lots of SPARC servers, and they are still used in some weird embedded applications as well. There are even new chips still in development. Definitely going out of style though.
To add to the trouble, Illumos doesn’t have much support for post-Sun SPARC hardware anyway. The codebase branched off of OpenSolaris in 2010, so it’s mostly frozen at whatever was supported then. As far as I know, there hasn’t been much (any?) interest in trying to get it running on newer Oracle SPARC hardware. The commercial users of Illumos (e.g. Joyent, OmniTI, Nexenta) have used it only on x86, and SPARC hobbyists aren’t buying recent $100k Oracle servers.
Which embedded ones are those?
This one! But to be clear it would never run Illumos anyway.
I have to wonder: How much of the userbase uses SPARC?
I postulate it has got more sparc users in terms of relative userbase than, say, Linux or netbsd. They may be abandoning half of their userbase.
Or maybe my guess is totally off.
Perhaps half their user base but evidently not half their developer base. If the user base ratio really is that high then more of those users need to step up to support SPARC at a developer level
It’s difficult to get an exact estimate, but I seriously expect the number of SPARC illumos users in 2021 can probably be counted on both hands. There’s only one seriously polished distribution for SPARC systems of which I’m aware, which is Tribblix. Over the last ten years periodically individuals will emerge that say they want to add SPARC support to some other distribution, but inevitably the effort goes quiet after a while.
The reality is that the SPARC systems we have a hope of supporting are not high performing by modern standards, so building a distribution worth of software takes a long time just to run the builds – and that’s assuming the software all builds already which it likely does not without patches.
Peter intends to manage a fork for SPARC support in Tribblix, so users of that distribution will likely be no worse off than they are today. I don’t expect this decision will impact more than ten or twenty people at most. Most all of the rest of the user base is using x86.