[OpenAFS] ...Those That Help Themselves

Derrick J Brashear shadow@dementia.org
Wed, 30 May 2001 15:04:02 -0400 (EDT)

On 30 May 2001, William Setzer wrote:

> After our own bad experience, reading the number of horror stories
> being posted to these lists, and hearing from other Universities with
> similar tales, it has become fairly clear that IBM doesn't really care
> about AFS.

That's not entirely clear to me, but we'll go on from here since I want to
comment on the rest of your message.

> and we are unlikely to see things getting better.  This is
> a big problem for NCSU, as AFS is a core component of making our
> academic environment work, and there's really nothing available with
> which to replace it.  I am certain that many of you are in the same
> boat.

Carnegie Mellon Computing Services decided several years ago that DFS
wasn't useful and there was no other feasible replacement. While AFS is
less a core service than it used to be because we have an independant mail
service now, it's still as important as ever.

> As such, my boss wants to me to feel out the other Universities and
> businesses.  Have you thought about what's going to happen to AFS in
> three years or if it'll even exist then?  What about a possible move
> to OpenAFS?  What are you going to do about it all?

We're moving on a path to OpenAFS; Our next-generation platform ports (at
this point, only Solaris 8) will start off using OpenAFS for clients, and
I'm in a position to switch our older Linux and Solaris ports to OpenAFS
as soon as I have some time to deal with a few very-site-specific local

> Perhaps it's time to band together and decide the future for
> ourselves.  OpenAFS is a very attractive option, but like all Open
> Source projects it relies upon dedicated volunteers and a decent
> install base to flourish.  NCSU is willing to dedicate some manpower
> to OpenAFS; maybe others will, too (if not doing so already).  I
> hesitate to call for the forming of a Consortium--we all know how that
> sometimes turns out--but something a bit less formal might be a good
> idea.

To some extent that's already happening; We have a road map of goals,
perhaps not as well delinated or distributed as it should be, and a lot of
it is just waiting for a developer or developers to pick it up. We have
people who are contributing as they can in various veins now, and I've
been doing integration, testing and distribution as necessary. To the
extent that that sucks up my available free time and the time my employers
are willing to fund for me to work on OpenAFS I haven't been able to do as
much development as I'd like, nor to have as many of the support services
for collective development available as I'd care to have done. 

Right now we need not just people interested in developing new features,
we need, essentially, port masters. We especially need Windows-centric
developers, but it looks like something is happening on that front. But we
also need people who are willing to test and build for some of the other
platforms; Currently I've been doing all the builds for Solaris, AIX,
IRIX, Digital Unix, and 2 of the 4 Linux distributions we support, and
hopefully we'll have HPUX support again in the near future. If we could
get a person or people to take on some of these platforms, it would be
very useful. I have no problem continuing to build and test for Linux
or Solaris as I have a large amount of resources available for those which
makes it easy to do; For AIX, though, I have a single machine, and for
IRIX and Digital Unix I'm borrowing cycles on remote hosts, and efforts of
other people who may have other priorities to test the build results. 

> Is anyone else interested in exploring these possibilities? 

I'm willing to work on necessary organization and software support for
better coordination of distributed effort, but at this point the first
logical breakdown of effort would be getting someone for at least IRIX,
Digital Unix, and the 2 Windows ports who are willing to coordinate work
for those platforms, and go from there. OpenAFS is on the cusp of
something very useful; After years of lacking ports for platforms which
were growing in popularity there's now client support for Windows 95/98,
Windows NT/2000 and Darwin/MacOS 10 (even if the MacOS 10 support isn't
very refined yet), which means there is finally the potential for AFS to
truly be everywhere. It's important we not squander this opportunity to
spread the use and usability of AFS.

Derrick J Brashear
speaking only for myself