> About the "obsolescence": There I have no great
> problems,
> because I read, hear and "know" that the free
> operating systems 
> today can handle the actual hardware and also much
> older
> hardware. Many businesses have chosen Linux because
> they could
> then use some older PC for "new" use cases without
> buying new
> HW. 
> 
> Maybe, Sir, I don't understand your obsolescence
> problem.
> Then explain it to me with some examples, please.

These OSs _now_ support the hardware that their maintainers choose
to support, including to varying degrees, older hardware.  But within
10 years, today's new hardware will be old indeed, both in terms of
performance, and in terms of poor energy efficiency for a given
performance level.  Some used hardware that's very well built will
still be around, but even if the price is low-to-free, with rising
electricity prices, it might not even be worth having.

Further, there will always be new peripherals (that people will want to use),
requiring new drivers. It wouldn't be so bad with Solaris (unless it moves 
mostly
to a different CPU architecture, like 4+GHz PowerPC or some such) inasmuch as
Solaris's interface between drivers and the rest of the kernel is _relatively_
stable.  But even that can't stay still forever.  And on Linux, where they
believe that source code means they don't need stable driver interfaces,
they make wide-ranging changes fairly often, which means that everything in-tree
gets updated, and new drivers have changed interfaces and different
kernel facilities to call on; thus, on Linux, backporting new drivers to
an older kernel could be relatively difficult.  Not sure what the situation
would be with the *BSDs.

And then there's CPU support.  CPUs within a family stay fairly compatible
at the unprivileged (user space) level, but less so in privileged mode; then
there's newer CPUs that have a hypervisor mode as well.  Just about every
CPU has _some_ bugs, that the kernel needs workarounds added for.  The
above distinction between unprivileged and privileged compatibility is probably
even stronger for SPARC, which gives itself more freedom by only defining
unprivileged behavior as standard; the privileged mode OS behavior has a
lot more latitude to evolve.  So eventually, you need to start backporting
CPU-specific support, or you can't run on new hardware anymore.

So, over time, you frozen (except for bug/security fixes) kernels would
_only_ be able to run on hardware no longer in production, and on some
of the OSs, would need more effort to have any new drivers one wanted
to add backported to them.  The effort would increase over time, and
the support problem for the old hardware would also increase, until spare
parts were no longer available, and people would have to try to scrounge
3 or 4 old systems to cannibalize for parts to keep one of them running.

Of course, just as an exercise, one could always run this stuff forever on
an emulator.  Heck, I've run Windows 3.1.1 on qemu, MVS 3.8j on Hercules,
CP/M on another emulator, v7 Unix on a PDP-11 emulator, etc.  Fun for
nostalgia, or if there was some particular app on one of them that was closed
source but did whatever it did really well.  But not particularly useful 
eventually.

Then there's the course aspect.  Over the last few years, we've had considerable
growth in virtualization, advanced filesystems and debugging technology, etc.
Something "must have" with design principles that students would need to know 
about,
is bound to come along in the next few years, and you won't have it.

I give this 6 months to 2 years to get set up, and no more than a 10 year 
lifespan before
it's of little further use, unless you allow for at least occasional (and 
perhaps synchronized)
updates of the kernels.

> > The cooperation sounds interesting and possibly
> very
> > worthwhile, although given the possibility of
> > what amount to planned obsolescence (2nd point
> > above), I wonder if there would be enough
> > interest to achieve much there.  Similar point on
> the
> > promotion of heterogenous environments.
> 
> The cooperation part is really new. The Unix guys
> have
> real problems with working together.
> 
> I see it as my biggest challenge to bring together
> the 
> OpenSolaris devs with the *BSD devs with the Linux
> devs.
> 
> But I want accomplish that. And I will be successful,
> I promise.
> Just now, I don't know how to overcome the "hatred"
> in the
> hearts of the users and devs alike for the other
> Unix-like OSes.

I don't think there's so much hatred as all that; the worst
of them, whichever that may be, still has much more
potential than Windows.  I do think there's arrogance
all around, plus some real anger at the fanboys in
certain circles that just _have_ to believe that they'll
rule the world one day.  _Nobody_ should end up ruling
the world, nor should there be just one community and
one culture; we'd all be the worse off for the lack of competing
ideas.  The competition however needs to be nonviolent (assuming
the others oblige - nothing says you have to let yourself get killed!),
and civil enough to keep the door open for cooperation whenever there's
some benefit to be gained from it, and keeping in mind that the
objective is _not_ to merge into one mind inhabiting multiple bodies.

> > A long, slow kernel update cycle (with snapshots
> of
> > the then current kernels being used as a
> > starting point only once the first multi-kernel
> > distro is out and stable) might address the
> > obsolescence issue, while still honoring the other
> > concerns.  It would also ensure that the
> > opportunity for improving cooperation would be in
> > depth and open-ended.
> 
> I want to start a "backporting" of actual drivers to
> the 
> "kernels" existent in Menhir. 
> With a "frozen" base system the testing will be easy
> I think
> AND
> The real interests in (home) users and developers are
> in
> application land today I think. 
> So with this I can create maybe a stable development
> platform
> for actual applications. 

All that still holds if you're willing to update the kernels every
four or five years, say; at which time you can still do minimum
essential maintenance on the old ones.  In other words, update
just often enough to maintain interest, relevance and usefulness.
It may turn out that could be never, but I really don't think it will.

> > I think that if you could engage as many Austin
> Group
> > (http://www.opengroup.org/austin/)
> > participants as possible, you could at once work
> for
> > more uniform standards compliance among
> > heterogenous implementations, and yet also for
> more
> > dynamism (albeit the responsible sort) in
> > the standards process itself.

I do think that relating this all to formal standards is
reasonably important too.  Interoperability should promote
cooperation, especially if the application space is the main
place for new development.  There are some difference
between for example Linux and Solaris as to how important
their individual communities are likely to consider formal
standards.  But the standards tend to allow enough latitude
that compliance shouldn't be unduly burdensome for any one
OS implementation, since after all, they are arrived at by
participation of multiple vendors.  So far, most of the
participants in formal standards have been commercial; but
this might be an opportunity to bring community based
development into the standards process in a big way.

The meeting of the disciplined process folks with those that
primarily believe in challenging every assumption could be
very interesting and beneficial.  Although to give some credit, both are
in the long run necessary, and those camps originally favoring
one and neglecting the other have both moved more towards
recognizing the value of both; witness the development of the
LSB, hardware vendors writing two-part drivers to isolate the
closed or model-specific parts from the OS interface, and so
on; and on the other side, Solaris incorporating features
such as zfs and dtrace, both of which (and especially the
latter IMO) challenge earlier assumptions of how things are
done.
 
 
This message posted from opensolaris.org
_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to