Re: Future of m68k - Etch and beyond
Ingo Juergensmann wrote:
> > On Tue, Feb 27, 2007 at 10:48:25AM +0100, Christian Brandt wrote:
> >
>
>>> >>> OTOH, I would be willing to invest more machines, time and money in keeping
>>> >>> the m68k port alive - regardless of being it a part of Debian or not.
>>> >>> Meaning: with all those political background Debian incorporates, it may be
>>> >>> worth a thought to fork the infrastructure and do our own stuff.
>>> >>>
>>>
>> >> To be honest Debian on m68k always felt bulky, overloaded and missing
>> >> the point. Give me another distribution which focuses on delivering a
>> >> lean core with at least make/gcc or maybe a simple package system like
>> >> ipkg and I'll bury my Debian-m68k-CD deep enough to never bother me again.
>> >> apt-cache search whatever takes several minutes on my Amiga 3000.
>> >> apt-get install can take hours even for medium packages. Before
>> >> Debian-m68k I had early experiments with NetBSD and some
>> >> Micro-Distribution (Watchtower?) and those distributions did not only
>> >> outperform in package management, in most cases their
>> >> memory-requirements were lower, their setup was more straight forward etcpp.
>> >>
>>
> >
> > I agree with you that dpkg and other tools are dog slow. I got the feeling
> > that especially scripts are very slow on m68k (interpreted languages).
> > The performance difference between AmigaOS and Linux is enourmous, even when
> > you consider the overhead of memory protection and other stuff. And the
> > difference is getting bigger and bigger.
> >
> >
>
>> >> I hate to be the first one to stop kicking the dead horse but, ya see,
>> >> its dead.
>> >>
>>
> >
> > The Amiga is told to be dead since the demise of C=, and even longer... ;)
> >
> >
>
I've run Potato, Woody, and Sarge on Macs ranging from a IIci to a
Quadra 840av, with Q700 and Q630 to fill in the middle. I've also run
Linux of various different distros on all kinds of x86 hardware from SLS
(anyone remember that?) on a '386/sx-40 to Ubuntu and Fedora on Athlon
systems. I used RH5.2 on a laptop with a 25 MHz '486 and 20M of RAM,
running Netscape 4 and WordPerfect 8 under FVWM95, which is rather
similar in capability to a Q700. Recently I've been using DSL quite a
bit on low-end Pentium systems.
I'm an old computer junkie. I love seeing what can be done with older
machines that others have given up on. I have lots of fun with Tandy
Color Computers and Amigas. Part of it is just nostalgia. Part of it
is a perverse desire to demonstrate the wrong-headedness of conventional
wisdom. Part of it is a result of being sickened and amazed by the
things that people throw away, like the 1G Celeron systems I dug out of
a dumpster.
But I also believe that there are some real, practical reasons to want
to continue to use older computer systems -- sometimes re-purposed to
play a different role than that originally intended, like the '486
running LEAF that shares my DSL connection with my household LAN. But I
think that there's still some good use in some of these old computers,
and I would be very disappointed to see all Linux support for the m68k
platform come to an end. Here are real uses I can imagine for m68k
systems of various sorts: (Some applications not suitable for all
hardware...)
1. Special purpose computer: firewall, network appliance, X terminal,
home automation controller, print server, file server, slideshow kiosk,
art installation, etc.
2. General purpose PC for lightweight GUI apps: word processing with
SIAG Office, FLWriter, or AbiWord, email with Silpheed, light Web
browsing with Dillo, and so on.
3. Programming platform for students, and amateur programmers like me.
M68k is great for learning programming: the assembly language is nice,
and there's plenty of improvement left to be made. There are a lot of
things somebody could contribute to open source operating systems for
the 68k. Not everything has already been done.
4. Increase the 'genetic diversity' of the 'Net. With Apple moving to
x86, we have reached a historic low point in the diversity of computing
hardware on the Internet. While the short-term benefits might be in the
economies of scale, one major down-side is that it makes the job of
malicious programmers that much easier. It's quite conceivable that a
single binary could be devised that would infect MacOS and MS-Windows,
and probably also x86-based Linux systems as well. But making it run on
a 68K Amiga or Atari would be much more difficult. And why would
anybody ever bother?
5. Many of these computers are free, or next to it. I've got two
Q840av's, 4 or 5 Q700s, 2 or 3 IIci's, a IIcx, a couple Mac IIs, a
couple Q630s, and probably more that I'm not remembering. I paid $5
each for the Q700s and one of the IIci's. The rest were absolutely
free. I spent a few bucks here and there on ethernet transceivers and
memory. Many of the people I build computers for say they really only
want them for word processing and email. If they could have a cast-off
computer that would do what they want for nothing, or a few dollars, why
should they have to spend hundreds?
I'm convinced from what I've seen, that m68k systems should at least be
capable of this level of service. I would simply like to see a
Linux-based, free OS for the Atari, Amiga, and Mac that would provide at
least the same level of functionality that these systems had with their
original, proprietary operating systems and application software. It
would be very nice if it could also remain under development, with the
focus being bringing compatibility with current standards vis a vis the
Web, word processing file formats, and so on, within the limitations of
the hardware, and with a simple GUI.
I don't see much point in spending a lot of effort trying to build GNOME
or KDE for m68k. Maybe XFCE. But I do see considerable value in
carefully selecting a subset of Debian packages that run reasonably well
on common m68k hardware and spending some time developing a distribution
based around them, possibly with a new, lightweight installer.
As I've said before on this list, I think Damn Small Linux would be a
great model for an m68k-centric mini-distro. It has Debian roots, which
should simplify morphing Debian/m68k in that direction. The MyDSL
mini-package system is great for small systems. And focusing the GUI
elements around FLTK would be a good move.
As Ingo J. mentioned above, the extensive use of script languages for
GUI stuff and installation doesn't make for a happy user experience on
the m68k platform. But DSL uses LUA with FLTK bindings, for things like
the MyDSL control panel. It seems to perform quite nicely on low-end
systems. I think that concentrating developer effort on things like
that could go a long way toward improving the usability of Linux on 68k
machines.
If the 68k platform is dead, maybe it can live on in some ghostly
afterlife. I envision a small, Debian-derived distribution centered
around the m68k. Maybe merely hundreds of packages isntead of
thousands. Call it Ghostship. Imagine a logo of a ship with sails
vaguely reminiscent of the old Motorola Bat Wings... It could include
elements taken from DSL (Dillo web browser patched for frames and tabs,
the MyDSL package system, the GUI control panels, the selection of
packages...), other Debian packages that are well suited to the m68k
plaform, and perhaps some older programs that have been dropped from
current Debian, but which could be resurrected. Like DSL, the whole
endeavor would be focused on packing the most real functionality in the
smallest memory and hard drive footprint possible. Perhaps the project
could later be expanded to include support for really low-end x86
systems (386-486).
I think such a thing is quite doable, considering the level of effort
that has already been put into continuing to support the m68k
architecture in Debian. Perhaps the time, computing resources, and
expertise would be better spent if it were directed toward a somewhat
more narrowed goal and a significantly narrowed set of packages.
If you look at the old, outdated registry of how many people have
installed Linux on m68k machines at some time in the past, it's clear
that if there were a distribution that would install a simple, usable
system on a 68k machine in an hour or two, featuring modest but
functional software, stripped of extraneous junk, like DSL is, there is
a significant potential user/developer base. If this system has a way
for users to install and build individual Debian source packages, then
they would still have at least some connection to the thousands of other
programs not maintained as an official part of "Ghostship".
The way I see it, the top development priorities for a project like this
would probably include:
*EMILE -- The only reason I don't run Debian on an Amiga or Atari is
that the requisite CPU accelerators still cost money (if they're
available at all) and computers with appropriate CPUs are comparatively
rare. Properly equipped Macs on the other hand, are practically falling
out of the trees. Getting EMILE to the point where one could boot an
installer from floppy or CD and run the system without MacOS would
dramatically increase the number of systems capable of easily installing
and running "Ghostship". From what I've seen, EMILE is getting close to
that point. There's just one piece of Apple software that needs to be
replaced in order to able to boot straight into Linux with no
proprietary code, IIUC. I used a test installer image mentioned on this
list that used EMILE to boot straight off CD and installed Debian on a
Quadra 630. So close. Too close to stop development now!
*Continued Kernel development -- pretty self-evident. Better support
for existing hardware in recent kernels.
*A simple installer. -- Why not just start with something dirt-simple,
like a very basic system image in a tar file that can be unrolled onto a
small hard drive partition like DSL. It doesn't even have to include
all the stuff DSL has, like Firefox, SIAG, VNC Viewer, RDesktop, XPDF,
XMMS, etc. That should bring the size down even below the 50M of DSL.
Other goodies can be added later using MyDSL, or something like it.
Maybe the installer could allow the user to select from a collection of
pre-compiled kernels stored on the installation medium. Whatever user
interaction there is should be done with programs that run fast on the
intended targets. The installer should be written in C/C++, or
Lua/FLTK. The latter would be a good compromise between ease of
development and performance, and has the added benefit of being able to
steal directly from DSL.
*Concerted development of carefully selected core applications. -- Pick
the best 2 or 3 window managers as measured by
performance/size/features/current development. Select apps using FLTK
and, secondarily, GTK+. Include fast, functional file managers, like
EmelFM, ROX Filer, and Endeavor 2. Put in a couple good image viewers.
GQview does the job well on my Q840av. Put in SIAG Office. It's really
quite nice. When you disable smooth scrolling, AbiWord isn't too bad on
older systems. Ted is pretty good as a rich text editor. Silpheed,
Dillo, etc. Put some effort into contributing to the development of
these projects. In some instances it might be useful to dig out older,
more svelte versions of some of these programs, and use them as 'lite'
versions for the 68030, or whatever. Possibly back-port newer
functionality to older, less eye-candy-heavy versions of some programs.
Maybe port some programs to FLTK that currently use other, less
well-suited toolkits.
*Port technologies that could improve the utility of m68k systems. -- At
the speeds of most m68k systems, just running X sometimes takes
significant CPU time, not to mention RAM. On systems like the Q700, you
can't get more than 20M in there unless you have nearly-mythical 16M
30-pin SIMMs. But as cheap and plentiful as these systems are, why not
use one as an X terminal and a second one to run apps? LBX and LTSP
would make this feasible. But I don't think there's any support for
using m68k machines on the terminal end of LTSP, though I think you can
use Mac/PPC. A Q700 w/2M VRAM would make a nice little X terminal. I
think this may just be a matter of building some source packages. A
more ambitious project would be to add support for something like
OpenMOSIX to the m68k kernel. A cluster of 68K machines isn't going to
make a supercomputer, but if you hang 3 or 4 Q700s off an 840av, you
might be able to have a few more processes running before you crawl to a
halt. Could this be useful for buildd?
I would prefer to see Debian continue to support the m68k architecture.
But if it simply isn't going to happen anymore, I really think that a
more focused effort, like what I have described, could result in a much
more usable, functional product, and could really justify all the
countless hours that have already been spent kicking what some have
called a dead horse.
JCE
Reply to: