Ubuntu 22.04 LTS To Shift Its PPC64EL Baseline To POWER9 CPUs, Dropping POWER8

[schwarzygesetzlos]

Hooray to Gentoo, Void, Adelie & Debian which continue supporting (older) PowerPC! \o/

Gentoo and Adelie support any 32bit/64bit PowerPC the Linux Kernel supports (>=PowerPC 603+?)
Void does ppc64 (>=POWER4 + Altivec) and any ppc32 (>=PowerPC 603+?) at the moment, but they will drop ppc32 and ppc64 below POWER8 until 2023 unless a maintainer steps up (https://voidlinux-ppc.org/news/2021/11/big-endian.html).
Debian does ppc64 (>=POWER4) as "release architecture" but 32bit ppc lost "release architecture"-status, but still is availiable through ports.

[Developer12]

It is indeed an EXTREMELY aggressive deprecation schedule, on par with dropping anything older than the last 3 or so generations of intel. I might expect fedora to do something that crazy, but not Ubuntu. This isn't even going to benefit POWER9 all that much, being as they were already tuning for it in the first place.

Add to that the fact that IBM still supports POWER8, Ubuntu's own testers seem to be using POWER8, and the change is only a couple of compiler options with no maintenance burden.

This really makes no sense to me and I'm suspecting it might be an internal mistake. Hopefully anyway. It might not affect me personally but it's going to cause a lot of pain among POWER users generally.

EDIT: It's also worth noting there's still a LOT of enterprise POWER8 out there, even if it's getting quite long in the tooth. Those orgs aren't going to be running Ubuntu Server anymore.

[kylew77]

In my opinion Ubuntu does this wrong. I think they should say the next LTS will be the last supporting LTS, but instead they say the last LTS will be the last supporting LTS. Announce it ahead of time not at current. They did the same thing with i686. Think people would take the Power 8 depreciation better if they said 22.04 LTS will be the last LTS to support Power 8 instead of saying it will be the first to not support it.

[tildearrow]

ppc64el? What do you mean? Like POWER without floating point?!
Like armel is ARM without floating point?

[chithanh]

PowerPC is largely an enterprise only architecture, not a lot of end users will be affected by a change like this.

Both you and Canonical are forgetting home labs here. Lots of decommissioned Enterprise hardware finds its way to home labs, which are often operated by admins, developers and semiprofessionals, and used as staging ground for getting familiar with technology.

If organizations absolutely need Power8 support for these new operating systems, they need to make their voices heard now (though it's probably a bit too late).

The interest among Canonical's paying customers is probably low, which is why they plan to drop it. I think this move will hurt POWER ecosystem most, as this will take out a significant chunk of its mindshare.

Red Hat has made this call earlier

Not surprising. Red Hat is mostly ignoring the home lab community now. They had lots of CentOS installs there, and when they decided to essentially discontinue that distro's popular flavor and instead offered a no-cost RHEL licensing scheme that was too limited for home labs, they largely burned that community. Some have switched to other RHEL clones, many have switched to Ubuntu.

https://www.servethehome.com/ibm-red...subscriptions/

[Dawn]

ppc64el? What do you mean? Like POWER without floating point?!
Like armel is ARM without floating point?

ppc64el just means little-endian byte ordering. ppc64 is big-endian. There are, to my knowledge, no PPC64 devices with no floating point.

armel originally just was to distinguish from the long-dead big-endian ARM port, "armeb", at a time when neither had floating-point. "el" has nothing to do with FP on its own - "armhf" was just newer and got its own name as a result.

[CommunityMember]

Wow they don't even attempt to give a rationale for a change that will seriously impact some users. That's certainly one way to treat your community.

Are you a *paying* customer? If not, your opinion is noted (and binned). Seriously, Power is primarily an enterprise target (those with handouts and discards noted), and only those contributing to Canonical's support by doing the (hard) work of full development and QA, or paying for that support, should expect Canonical to give a darn. TANSTAAFL, and should Canonical wish to survive (and if you wish Canonical to survive), that includes Power support.

[WorBlux]

PowerPC is largely an enterprise only architecture,

Power ISA is now open, so new embedded or mobile class products showing up wouldn't be a huge surprise.

Tying the power c library and compiler so heavily to the hand optimizations made for power9/VSX is not only a hurdle for legacy support, but makes it more difficult to actually bring up more mainstream products.

[mroche]

Both you and Canonical are forgetting home labs here. Lots of decommissioned Enterprise hardware finds its way to home labs, which are often operated by admins, developers and semiprofessionals, and used as staging ground for getting familiar with technology.

The interest among Canonical's paying customers is probably low, which is why they plan to drop it. I think this move will hurt POWER ecosystem most, as this will take out a significant chunk of its mindshare.

Not surprising. Red Hat is mostly ignoring the home lab community now. They had lots of CentOS installs there, and when they decided to essentially discontinue that distro's popular flavor and instead offered a no-cost RHEL licensing scheme that was too limited for home labs, they largely burned that community. Some have switched to other RHEL clones, many have switched to Ubuntu.

RHEL 8* and derivative projects choosing to match will be supported until end of May 2029 (2031 for ELS subscribers). Customers of Canonical's ESM program will have until (end of?) April 2030 with Ubuntu 20.04 and non-customers until 2025, should this change not be reverted. Owners of Power8 hardware can still use a supported version of an Enterprise Linux family distribution or Ubuntu until those points in time, just not the newest and shiniest. It's not as if when RHEL 9/Ubuntu 22.04 comes out Power8 support is going to vanish. Hardware cycles are a known occurrence in the enterprise space; not everyone likes them but the become necessary as time moves on depending on install base trends, performance and optimization, and support burdens. These trends come from existing customers and potential business opportunities, not the home/after market which is relatively trivial in comparison and not either company's business focus.

Enterprise groups do _not_ move quickly from the foundational software perspective, newer hardware is often purchased well before operating system major upgrades occur. And they definitely don't jump when a version is first released, often 2-3+ years into its cycle. That hardware won't go bust when it enters the aftermarket, currently existing distribution versions will still handle it fine.

* Though oddly the ppc64le build is not available for download in my personal account using the Developer Subscription for Individuals. I might post this internally next week and see if anyone bites. The RHEL 9 beta version is, but that doesn't help this conversation.

Cheers,
Mike

[Space Heater]

Are you a *paying* customer? If not, your opinion is noted (and binned). Seriously, Power is primarily an enterprise target (those with handouts and discards noted), and only those contributing to Canonical's support by doing the (hard) work of full development and QA, or paying for that support, should expect Canonical to give a darn. TANSTAAFL, and should Canonical wish to survive (and if you wish Canonical to survive), that includes Power support.

Ah yes, if Canonical had an employee spend 10 minutes writing a few sentences explaining their decision rather than just declaring it, Canonical simply would not survive!

Their behavior here is far below what many other companies do for non-paying customers, not sure why you're trying to defend it.