PowerPCReview

Differences between revisions 31 and 34 (spanning 3 versions)
Revision 31 as of 2007-01-31 00:12:04
Size: 23372
Editor: ip72-204-230-239
Comment: Mark's Longish Comments (typos fixed)
Revision 34 as of 2007-03-05 20:10:22
Size: 15695
Editor: cpe-76-173-8-128
Comment: delete spurious discussion, have emailed author
Deletions are marked like this. Additions are marked like this.
Line 139: Line 139:
==== Mark's Longish Comments ==== ==== Linus Torvalds uses PowerPC ====
Line 141: Line 141:
(Admin: if this entry is mis-formatted or mis-positioned feel free to change.) Here's a news article of some relevance from 2005/03/09:
Line 143: Line 143:
PowerPC is a learning opportunity. In 2 years the next great shining chip will emerge. It will happen regularly. And OEMs will switch horses mid-race. And use black boxes for customer lock-in. And release multiple variants of the same model. Welcome to consumer electronics. [http://www.builderau.com.au/news/soa/Linux_creator_Torvalds_switches_to_an_Apple_Mac/0,339028227,339180280,00.htm Linux creator Linus Torvalds is now using an Apple Macintosh dual G5 as his main desktop]
Line 145: Line 145:
In this hurly-burly Ubuntu proposes 5-year support programs. The issue is how to map it across fickle and sometimes evil OEM behavior.

The issue is general so frame a general question. Simply "getting rid" of today's problem chip will not solve systemic problems.

Canonical might need to fix its work flow. Starting from an assumption that current work flow is correct leads to false conclusions. The core issue raised is "if we are going to apply paid developers to this port." The implicit fallacy is that Canonical must continue in the same (frustrating?) fashion.

The decision in the instance may be right. It could also be the wrong focus, relieving a symptom while leaving the disease to manifest another day.

Build System Adaptability

Examine build system adaptability. T2 Linux runs flawlessly on PowerPC, both 32 and 64 bit. T2 claims that adding a new architecture can take just one day. How long does it take Ubuntu? http://www.t2-project.org/about.html

Test and QA and Release Festivities

Ubuntu enjoys thousands of users willing to test alpha software. Most companies chase bugs on their own, because code is proprietary. Ubuntu enjoys practically unlimited testing labor. Ports in this context should be easy.

The way to exploit this pool is to isolate bugs rapidly. Swift isolation allows developers to focus energy. Multiple reports confirm where bugs live. Here Ubuntu does pretty well. Bug reporting systems are working fine. Users are doing their part.

The problem is follow-up. Confirmed bug reports sit unattended for months and even years (i.e. across final releases). I could cite examples from PPC G5.

Ubuntu enjoys something else that most companies don't, free development labor by Debian, which supports PowerPC already. And Ubuntu supports it too. An economist would call that sunk cost.

So the Canonical situation is vastly different from a standard enterprise. Ports of this type should not be so difficult. That they are considered so indicates problems in work flow. I would not say that of any other enterprise in a similar situation. But with all the users, plus Debian helping, for free, I am more than willing to offer this constructive critique.

Those willing users will gladly QA your hardware over the net (install testing etc.) and relieve Canonical of awful chores. The converse is possible - users volunteering their own hardware to Canonical for test, over the net or physically. I've seen it happen on Ubuntu boards.

Test installs can be performed by volunteers with only occasional intervention by BenC and co. Volunteers will even show up on Canonical's doorstep, perhaps asking merely travel expenses.

Think of it this way. Industry uses pools of "temp labor" for high-demand seasons like Christmas retail and autumn harvest. Consider release dates as seasons to get temp/volunteer labor along these lines.
(To BenC: it's not "just" a matter of test installs, but they do count.)

Hardware

Bugs in ports will relate to specific hardware. It's cheaper to buy hardware than pay developers - so buy hardware. The complaint arises, we have no office space. Once you fix hardware bugs, the unit can go into a closet, garage, or eBay. Or better, it can go to a back room for VNC use from the front office.

For example bugs in 64-bit G5 are all hardware, known and characterized, user reports filed for many months (even years). Most commercial firms would envy Canonical's position here. But in all that, Canonical has not obtained a single hardware unit (AFAIK) or worked attentively on these very narrow issues which, in isolation, are not that difficult.

Let's not blame one chip or platform for a general industry trend. The migration from 32 to 64 bit is going on everywhere. Some G5 bugs resulted from 32 vs. 64 bits, not PowerPC as such. Any architecture migration like that will have these glitches. There has of course been zero resistance to x86-64 work.

You can't buy every box in the universe, I agree. Decide what to buy this way: the free test labor pool indicates where the hardware bugs live. Buy those specific units and fix the bugs.

Canonical should employ 1-4 low-level kernel/hardware hackers who do nothing but this type of work.

Making Friends with Debian and Linux

Ubuntu owes a great deal to Debian. The preceding is an excellent way for Ubuntu to give back - solving hardware bugs in specific platforms.

Servers vs. Desktop

My view of the idea that supporting only servers will lighten the load, is that it's based on a wrong understanding of hardware. We're in the day of commodity electronics. In decades past there were major architectural distinctions between mainframe servers and terminals. Today there are not. It's more a question of tailoring.

Note one prominent example, Google, networking cheap commodity PCs. In fact an entire cottage industry of clustering is built on them. Currently, Linux runs a large percentage of embedded systems (some claim 50%), many of which are PC/104, another widespread cottage industry. These units are identical to desktop platforms except in physical specifications. They run the same code.

Commercial Viability of Canonical

A good marketing bullet is to say that Ubuntu will run on practically any hardware. The latest hot sexy CPU will change every 2 years. The PowerPC is not exactly fading, only Apple's use of it, while Cell looms on the horizon. This whole episode should be considered as a chance to make porting functional before the next wave hits.

The Charity and Third World Perspective

The third world often runs used or old equipment. If Canonical wishes to help this sector, then supporting what OEMs abandon for profit reasons makes perfect sense.

The Synergy Issue

Responding to BenC's "I notice a trend in comments. Try to understand people," let me say one thing. There is more at stake than crying users feeling abandoned. That's not the half of it.

There are zillions of "community driven" distributions. Some live, some die, and some limp. What made Ubuntu mindshare explode is public/corporate confidence in the corporate backing (employees vs. hobby hackers) of a give-away spirit (we'll mail CDs for free and work with the hackers). This synergy is not to be taken lightly. It is one of the driving forces in Ubuntu's current success.

Going into enterprise servers for revenue generation is certainly possible, but should not in the process destroy Ubuntu's unique contribution and position in open source. Personally I think Canonical will forge entirely novel revenue generation mechanisms based on this mighty synergy. Following the pack of enterprise server vendors is not the way to innovate.

The Intelopoly

Break the Microsoft Windowsopoly, yes. Let's not forget the Intelopoly. It is a problem in industry. You find Intel penetrating deeply everywhere, even the embedded market.

It is worth remembering that Intel CPU design is a mess. The mess runs faster and faster every year but the architecture stinks. It has hindered industry in the same way as Microsoft Windows, another broken design. To the extent that Ubuntu can assert influence here the greatest public benefit lies in sapping Intel's dominance.

Intel is combining with Microsoft, Hollywood, and Apple to enforce DRM, among other things, and Linus hates their EFI firmware. http://kerneltrap.org/node/6884 Meanwhile, open-source firmware is on the rise. http://www.mail-archive.com/devel@laptop.org/msg01849.html http://www.linux.com/article.pl?sid=06/11/30/199208

Please check the status of this specification in Launchpad before editing it. If it is Approved, contact the Assignee or another knowledgeable person before making changes.

Summary

PowerPC, already a significantly less mainstream architecture than x86(-64), has seen its visibility further reduced by the fact that Apple, the primary source of consumer PowerPC hardware, has moved away from the platform. Ubuntu needs to decide whether PowerPC should continue as a fully supported platform for the feisty release.

Rationale

  • Reasons against keeping PowerPC as a fully supported architecture:
    • In July 2005, PowerPC represented 1.95% of downloads from archive.ubuntu.com (this figure is more or less the percentage of the current user base performing upgrades). In only 15 months (i.e. to November 2006) that number has declined to 0.8%.
    • In January of 2006 Apple started switching from PowerPC to Intel and by August 2006 its entire consumer line had switched away from the PowerPC architecture. That leaves people with (as good as) no access to new consumer level desktop or laptop PowerPC based hardware.
    • PowerPC represented 3.3% of CD downloads from releases.ubuntu.com for dapper but only 2.7% of CD downloads for edgy.
    • Testing an Ubuntu release involves a complex testing process that currently involves testing 22 variations for each architecture. Fully supporting PowerPC means the full 22 variations have to be tested which is a strain on available resources especially since testing requires physical access to hardware, much of which is no longer on sale.
    • archive.ubuntu.com and releases.ubuntu.com are mirrored by hundreds of mirrors worldwide. Each architecture increases the total size of the archive and the size of updates sent to all these mirrors everyday.
      • Overall size of PowerPC on archive.ubuntu.com: 27Gb (15% of total size)
      • Overall size of PowerPC on releases.ubuntu.com: 9.2Gb (32% of total size)
    • PowerPC currently provides no production-level virtualization support to the degree we require for PPA (PersonalPackageArchives) or Grumpy. This makes it harder for community members without access to PowerPC hardware to support the port.

      • According to BenH, this is untrue. I'll try to verify on my OpenPOWER Box. The reason for Xen not being supported is that for the most part, IBM provides its own hypervisor in hardware (much like sun4v).
    • Several of the PowerPC specific tools (e.g. yaboot) are essentially unmaintained and trying to keep them (and the PowerPC) port current with new technologies (e.g. gfxboot) is difficult and time consuming.
      • Untrue: BenH: yaboot is actually maintained again (Paul Nasrat is the maintainer). There's also a port of grub2 under way.
  • Reasons for keeping PowerPC as a fully supported architecture:
    • When Apple EOL support for their PowerPC based machines, there'll be several knock on effects:
      • Existing Apple users may choose to migrate to Ubuntu so they can continue to use a supported OS
      • Existing Ubuntu users may be able to pick up second hand PowerPC hardware very cheaply
        • Both of these may lead to an increase in demand for the PowerPC port.
    • The new Intel-based Macs are [https://features.launchpad.net/distros/ubuntu/+spec/intel-mac-support not yet supported], so this risks leaving loyal Apple users out in the cold until that is fixed.

    • Sony's PS3 is PowerPC based and will support running Linux. It will be a modern and powerful platform, easily available for end users.

    • IBM's OpenPower platform ensures that powerpc remains a viable Linux platform on the server

    • There are some existing IHVs (e.g Genesi) producing powerpc based consumer hardware
    • [http://www.power.org/ power.org] exist to try encourage further IHV powerpc based consumer hardware

    • Xen is being worked on, however it requires hardware assist and only certain types of PowerPC hardware are suitable (notably, Apple Hardware and OpenPOWER servers are not).
    • [http://www.genesippc.com/products.php GenesiPPC] sold more pre-installed Ubuntu systems than Dell in 2005.

    • iMacs are a major platform used in schools; at least for Edubuntu and LTSP, PowerPC support should continue for a while.
    • Maintaining multiple primary platforms forces platform independence. Ubuntu is already down to only three primary targets, reducing that to x86(_64) only would encourage x86 specific design choices, which would bite us later.

Use cases

Scope

This specification is limited in scope in the following ways:

  • It only considers the future of the PowerPC port for the feisty release. This specification will need to be revisited for feisty+1.
  • Removal of the PowerPC port entirely is not under consideration.

Design

Implementation

There are three possible alternative implementations:

  • PowerPC remains a fully supported platform
  • PowerPC becomes a 'supported on the server only' platform (like Sparc)
  • PowerPC becomes a community supported platform (like PA-RISC and Itanium)

Code

Data preservation and migration

If PowerPC becomes a community supported platform and moves to ports.ubuntu.com, there will be some some migration issues.

  • Migration of feisty only for PowerPC to ports.ubuntu.com while retaining edgy and earlier releases on archive.ubuntu.com

  • Changes necessary to debian-installer and other packages to adjust for powerpc being on ports.ubuntu.com rather than archive.ubuntu.com

The former is complicated by the fact that the current archive/ports split is achieved by a simple list of rsync exclude and includes. If feisty PowerPC moves to ports.ubuntu.com, the split process would have to become significantly more intelligent (e.g. by parsing the Packages and Sources files to determine what files need to be copied across and/or excluded).

Unresolved issues

The fundamental decision remains undecided.

BoF agenda and discussion

["Warbo"]: Since 6.06 has long term support it might be a viable option to not port Feisty to PowerPC, since Dapper will still be supported. The server issue is similar, since Dapper is more of a server-focused release than Edgy or Feisty. Admittedly I think one of the great advantages of Free Software like Ubuntu is not being tied to a single architecture (x86), but with the current situation I think that only supporting PowerPC with LTS will produce more useful results for x86 users, and as PowerPC-based devices enter more and more homes (mainly as consoles) and GNU/Linux systems become easier to install and use on them then the situation can be revised to take advantage of this untapped userbase. Mind you, perhaps Ubuntu is not suitable for console systems at all and we should try not to let Ubuntu's influence and popularity stifle new distros with a narrower focus on that area.

["KeywanNajafiTonekaboni"]: I am a PowerPC User and find some points in your list against PowerPC-support not fair.

You point to the less downloaded CD Images of Edgy for PowerPC. I download breezy for my iBook one time. I installed Ubuntu and never again download a CD Image again. I am using Ubuntu Edgy now and uploaded every Version with apt-get. You also encoureged users to stay by Dapper, because edgy is more experimantel then the releases before.

  • The relative size of the existing user base is already addressed by the figures from archive.ubuntu.com. The CD image download figures were intended to address new installs only. I've clarified this above. --ColinWatson

The other "fact is, that you can't buy PowerPC Hardware from Apple anymore. You can still buy PowerPC based Hardware on eBay etc for testing (or using). Richard Hughes, the developer behind the awesome Gnome Power Manager, bought a broken iBook G3 "Clamshell" to discover the PMU (PowerManagementUnit) and support it.

Dapper is LTS, that's true. And edgy supports PowerPC, but I don't want to stick with GNOME 2.16. If I like that, I still would using Debian Wink ;)

Please don't drop the support for PowerPC. I think they are a lot of People using their iMacs, iBooks and don't like or have the money to throw they old computers away and buy new hardware. And it's just one year ago Apple dropped their Hardware. Can't you wait at least one or two releases?

And a last question: How much time and money you would really save, by dropping the support. 40GB HD space is nothing. 22 test sounds a lot, but are they automated etc. How much attention need this test?

  • These tests are presently entirely manual. We are working separately on automating some of them, but they still consume a very large amount of core development team time just before each release (including milestone releases, etc.). Furthermore, the lack of support for Xen on powerpc makes certain kinds of automatic testing more difficult. You brush off 40GB of disk space but it's actually a fairly major issue for some of our mirrors; note that Ubuntu shares space with other distributions on many mirrors and additional tens of gigabytes mean that they have to make hard choices and/or buy new disks for machines that are often already physically maxed out and cannot easily be upgraded without significant expenditure. --ColinWatson

  • I don't see how the 40GB can be a problem since mirrors have the option of not carrying a particular platform. If this is an issue with the official country archives, perhaps it would make sense to carry the ppc repositories only on the main archive.ubuntu.com servers and not on the country archives. Since the downloads are limited anyway it should'n cause unmanageable overhead to those machines. --EduardGrebe

Could an active PowerPC team, or PowerPC person be found to work on some of the specifically PowerPC issues in ubuntu. Launchpad lists a ubuntu-powerpc, but this only has 2 members (both of whom have plenty of other jobs to do). --SamTygier

One point made in the forum thread (see below), is that there are a few PowerPC specific bugs keeping people away from ubuntu. Maybe resolving these would increase the number of PowerPC users. --SamTygier

There are currently many questions surrounding the investment of time required, and it seems some people would be willing to give their time or help search for someone(s) to assist in continuing active support of this arch. It may be possible to offload some of the testing required to this (not yet assigned) person. If this is the case, the small overhead cost of getting someone involved is outweighed by greater ease of supporting ppc. --EvanMcNulty

I notice a trend in comments. Try to understand people, in no way is this spec about getting rid of the Ubuntu PowerPC architecture. No matter the outcome, you will still be able to get the latest and greatest Ubuntu release for PowerPC. What it is meant to decide is if we are going to apply paid developers to this port. This is more than a matter of harddrive space and doing test installs. If we decide not to support it, then it moves to ports.ubuntu.com, along with ia64 and hppa, and becomes community supported. It seems like we have plenty of volunteers to take care of it, which is what we would want. However, no decision has been made yet. -- BenC

But it does mean ending official support, BenC, which is what's most important to me. I think Ubuntu's PowerPC exit strategy is fundamentally flawed: For one, not even Apple has left the platform.

Please, everyone, take a moment to think back to the 68K to PowerPC transition. The first version of the Mac OS that didn't run on 68K machines was 8.5, release a full 4 years after the transition.

The same thing is happening during the PowerPC to Intel conversion: Apple's already stated that they'll continue to support PowerPC "for many years to come." Sure, they're not making anything with PowerPC processors but software development on PowerPC hasn't stopped: The current version of the Mac OS continues to receive security & other updates. In addition, Apple's already confirmed that the next major release of OS X (10.5 aka "Leopard"), due Spring 2007, will run on both PowerPC and Intel machines. Given Apple's past track record of supporting older hardware (think back to the 68K to PowerPC transition again), it's reasonable to conclude that 10.6 (and perhaps even 10.7 although that is less certain) will support PowerPC as well. So the folks at Ubuntu can't use "Apple's left the platform" as an argument to drop official support, IMO.

(rant)Shuttleworth always said he intended to make money from Ubuntu. I guess losing official support is the downside of using a distro what wants to be commercial: If your platform isn't considered big enough, official support is dropped.(end rant)

IMO, Ubuntu should continue official support for PowerPC for at least as long as Apple does, and perhaps longer. There is, for example, no way for users of PowerPC Macs to upgrade to Intel ones & continue to use Ubuntu. If official support is ended, Mac owners will have no official support at all.

At least work on getting Ubuntu to work on Intel Macs before ending official support for the PowerPC version. In this way, Mac owners will have the option to maintain official support.

There is mention of this issue being decided in December, but I can find no reference in the spec, this thread or the wiki page as to HOW it is going to be decided. Any guidance here?

MattZimmerman: this issue will be decided at the next meeting of the Technical Board (see TechnicalBoardAgenda).

- I agree with most of Canonical's mindset on this issue. As Ubuntu 6.06 LTS will continue to be supported until mid 2009, there will still be plenty of "official" (read: paid) support there. And new releases can become purely community driven. Yes, it does seem unfair to current PPC users at first, but in reality, these people will most likely see very little to no difference at all in their user experience. I also think that keeping official support for the server edition, while maintaining the desktop versions as purely community driven would also be a good move. The PS3 also presents an interesting situation, but I do not think it will be a problem here either, as Ubuntu will still be available just without Canonical's money behind it's developers/maintainers. -- Derick_eisenhardt 2006-12-25 @ 01:34CST

Linus Torvalds uses PowerPC

Here's a news article of some relevance from 2005/03/09:

[http://www.builderau.com.au/news/soa/Linux_creator_Torvalds_switches_to_an_Apple_Mac/0,339028227,339180280,00.htm Linux creator Linus Torvalds is now using an Apple Macintosh dual G5 as his main desktop]

Ubuntu Forum PPC Users discussion thread

http://ubuntuforums.org/showthread.php?t=298746

Petition to keep PPC

https://wiki.ubuntu.com/SavePPC


CategorySpec

PowerPCReview (last edited 2008-08-06 16:18:55 by localhost)