[9fans] Oil on the water

[James A. Robinson Jim.Robinson@Stanford.Edu]

> We are considering doing a new distribution of Plan 9, the system we have
> been calling Brazil internally to distinguish it from the released system.
> It's not promised or even decided yet, but given the recent discussions
> in 9fans it seems worthwhile to explain what we're thinking about.

It is good to read that it is at least a possibility.  Focusing on the
PC probably makes sense these days, since the hardware is cheap. If,
as I've read, Brazil has multi-processor support for x86 architecture
then it should prove plenty fast as well.

>From what what I've read on this list, it seems like there are some
differences of opinion between the researchers and the lawyer/management
folks at Bell Labs.  It's too bad that this happens -- I think Linux has
shown how useful an open policy can be for developing and enhancing the
value of an operating system.  The downside is fragmentation and what some
view as the "corruption" of the concepts behind the original design.

DigbyT brought up an interesting point: outsourcing Plan 9 distribution
to a real technical company might be a good way to generate money and
have that group handle the headache of distribution & installation
support. I wouldn't get our hopes up though -- I imagine that if Brazil
is released, it will need a lot of polish before vendors see a potential
source of revenue.

Of course, one idea might be to deal with Walnut Creek -- they have
an enormous FTP resource, and sell CD-ROMs of their archive at cheap
prices.  If Bell Labs would be willing to open Brazil up a bit and
allow folks to contribute/develop/polish Plan 9 into easy-to-install
form, it just might take off like Linux has.  Might that be worth
the initial effort from Bell Labs?


> The main obstacle to a release is pragmatic: the work of putting
> together a coherent package of hundreds of programs and the better part
> of a million lines of code is not something to be undertaken lightly.
> There were several man-years of documentation work alone in the last
> distribution.  We don't relish going through all that again.  Also, there
> are inevitable difficulties with distributing a bootable OS that do not
> arise when distributing a single application.  As you all well know,
> these were a major source of trouble with the last distribution.

Out-sourcing, letting groups of people make Plan 9 distributions, might
be  just the solution for this. The Linux Documentation Project has
made great strides, and the various distributions (Debian, Slackware,
Redhat, SuSe, etc.) have become very impressive compared to the original
installs I used (SLS).  SLS used to have sparce documentation, was hard
to install, and wasn't very user friendly. But thanks to the open policy,
there is now an enormous amount of high-quality documentation and there
are quite a few easy-to-use installations.

Course, the lawyers probably don't like that idea. It is odd -- shouldn't
it be obvious that if Plan 9 releases 1 and 2 didn't take off, that the
current scheme might not be the optimal way to get this research OS into
the hands of those interested in OS research? :)


> If we wait to get all those cool things done that make a distribution
> worthwhile for all concerned, you'll have to wait quite a while.
> Discussions continue on how and when to proceed.  Meanwhile, the system
> improves.

If you release something now, and then let people set up a system to
merge improvements back into the publicly available version, couldn't
your people then take the parts they liked and eventually merge version
4 into the public tree?  This is similar to how the Linux groups work --
not every modification gets back into the kernel, but the good ones do,
as well as any improvements from the dedicated kernel writers.  I think
the public might have something to offer back to Bell Labs, if we get
access to modern software as a starting point.


> Any release is likely to be done over the web, on a `fee-free basis
> for noncommercial use' (that is, no $350 price tag).  However, the
> necessity for negotiations for commercial use will almost certainly
> remain.  There will probably be no CD or other artifact, to keep costs
> down and simplify publication of updates.

I think the FreeBSD group has a large CVS tree that all their projects
reside in. They then have a web interface to this tree, which lets people
easily scan changes that others have made, and import those changes into
their own copies. Perhaps that might work for you folks?  Problem is,
how to justify the labor from a commerical organization's point of
view. But perhaps external contributions might make it worth that?
I imagine there must be at least some people out there who have good
ideas that would prove useful to Bell Labs development.


> Please don't get too particular about licensing details now; it's far
> too early and out of our (CS Research's) hands, in any case.

It's too bad you folks don't get more say about the external future of
the programs you brain-storm, design, and develop. :-(


Jim