Re: [Autoconf] Call for comments to a new AUTOCONF charter proposal.

["Charles E. Perkins" <[e-mail]>]

Hello folks,

I find the proposed charter extremely disappointing.
Comments below.

> Just so that you know who to blame :-) I was the one who asked Thomas
> and Ryuji to consider a simple solution with no new protocols.

If there's anything I've learned while working on
ad hoc networks, it was that existing protocols
just plain don't do the job.

> But back to your feedback on the charter. I would like to respond in two
> ways. First, I am by no means wedded to the particular solution details.
> Do you have a suggested edit?

I think we should consider solutions that have
a chance to work in general ad hoc networks, and
in my view DHCP is quite unlikely to do so in
any scalable way, for reasons that we've discussed
on the list seemingly dozens of times.

> But, second, I will defend the idea that the working group needs to
> learn to walk before running.

This analogy, if accepted as accurate, pretty much
ends the discussion.

But I do not think it is accurate.  We've had solutions
on the table for years now.  These solutions are known
to work.  There are multiple good ideas, based on
sound understanding of the problem, then built and
checked.

>                        We've been through a five year exercise
> just to define the addressing model. When we attempted to define the
> architecture as a general model we realized that it was hard, and
> explaining the model to non ad hoc networking experts was even harder.

The way I see it, we spent five years trying to stop
treating everything like an Ethernet.  I wouldn't have
wasted so much time on it, except that I thought the
result would be that we could then specify an address
autoconfiguration protocol that didn't pretend everything
was an Ethernet.  Maybe I was wrong, too wildly optimistic,
in that regard.


> But when we described a concrete addressing model that some deployments
> are using we did finally get an RFC. Or almost have, at least. I would
> like to continue on the same path. I'm told that there are deployed
> autoconfiguration mechanisms and that some level of support is doable
> even with existing protocols.

The addressing model as recently completed does
support a decentralized approach.

> I would like to avoid repeating the experience we had earlier on trying
> to describe the autoconf problem.

Didn't we finish doing that by now?  The next step would be
to specify an autoconfiguration protocol that could work in
ad hoc networks.  We have several candidates, and we could
compare them for applicability, performance, etc.


> .... But I do not look forward to more years of discussion of
> complex routing protocol/neighbor discovery extensions and a constant
> stream of questions from the outsiders about why ND or DHCP cannot do
> the job.

Ahh... I guess you mean that the outsiders are now
controlling the destiny of [autoconf] because they
can soak up so much time asking about why we aren't
interested in programming virtual Ethernets.

> I do get that a solution with a central node (and perhaps any solution
> with DHCP) does not solve all needs. Would you be happier if the charter
> had four work items:
>
> 1. Design space survey (Informational)
> 2. The simple solution (such as DHCP) (PS)
> 3. Limitations of the simple solution (Informational)
> 4. Recharter for work on the more general solution

I'd be happy if it were possible for [autoconf] to
be allowed to consider the excellent body of work that
was seen already years ago -- the same body of work
that motivated me and others to create and spend a lot
of time over the last years and years.

If I had known at the outside that the answers were
"DHCP" and "avoid constant questions from outsiders",
I would have had a lot more time at my disposal for
other things.

Regards,
Charlie P.