[Autoconf] Review of tentative draft-ietf-autoconf-statement-03.txt

["Ulrich Herberg" <[e-mail]>]

Dear all,

I have some comments on the autoconf statement (03). In some way I
agree with Dave Thaler (comments inline):

> > From: Dave Thaler <[e-mail]>
> > Date: November 21, 2007 10:34:11 PM CEST
> > Subject: Review of draft-ietf-autoconf-statement-02.txt
> >
> > 1) The document is supposed to be a problem statement, and it does
> > not really state what the problem is.  It concludes that existing
> > solutions can't work, but provides no discussion of this to
> > motivate that conclusion.  That is what, in my view, is the point
> > of a problem statement document and as such should be a significant
> > contribution.  This is the main reason why I feel it needs
> > significant work before WGLC.

In my opinion, the problem statement should state more prominently the
problems of autoconfiguration and existing solutions (e.g. IPv6
stateless autoconfiguration, DHCPv6). In section 4.1.1, part of these
problems are mentioned, but under the section "goals". I would rather
change the order of section 4.1 and 4.2 and first speak of issues THEN
of goals how to solve these issues. To my understanding, a discussion
of problems of existing solutions would rather belong to the section
"issues" than to "goals".
Maybe one could also underline that the main problem of existing
solutions like IPv6 stateless autoconfiguration is not only that it
"do[es] not account for potential address duplication beyond a single
[...] link" (from section 4.1.1), but that it has a completely other
intention: namely to configure HOSTS, while we are talking about
assigning prefixes to ROUTERs.

In section 4.1., "the primary goal of MANET autoconfiguration is [...]
to provide mechanisms for IPv6 prefix provision and address assignment
for operation on MANETs". This formulation -- though probably correct
-- sounds a bit vague for me. I have another idea in my mind which
would be:
"The primary goal of MANET autoconfiguration is to allocate globally
unique and topologically correct prefixes, and IP addresses to MANET
interfaces. "

This primary goal is probably not always desirable or necessary;
that's why I would add some "subgoals" in a hierarchical sense:
1. configuration of unique link-local addresses / prefixes
2. configuration of unique MLA addresses / prefixes
3. configuration of global addresses / prefixes
One or more of these subgoals can be achieved by autoconfiguration.

Looking at the table of contents of the draft, I see the following
subsections of
  4.1 MANET Autoconfiguration Goals
    4.1.1 Multi-hop Support
    4.1.2 Dynamic Topology Support
    4.1.3 Network Merging Support
    4.1.4 Network Partitioning Support

Are these really the the main goals of autoconfiguration? If I think
of goals of autoconfiguration, I would have something different in
mind. Some of these would be (feel free to comment...)
    - MANET nodes need to have unique prefixes within the desired scope
    - The prefixes of MANET routers should be easily aggregatable
using Classless
      Inter-Domain Routing (CIDR).
    - Hosts should be automatically configured using the prefix of the
router they are
connected to (e.g. using IPv6 Stateless Autoconf.)
    - All subgoals mentioned above should be accomplishable:
      autoconfiguration of unique link-local addresses/prefixes,
autoconfiguration of
      MANET-local unique addresses/prefixes, and configuration of
global prefixes.
    - The achievement of one subgoal should not depend on the accomplishment of
another (e.g. allocation of MANET-local prefixes should not depend on link-local
addresses).
    - The solution should be independent of any particular routing protocol.


Another idea would be to think about the following question: "what do
we actually want to autoconfigure? hosts? routers?". I would propose
add some sentence like:
"FIRST we need to autoconfigure the MANET interface of the ROUTERS
(i.e. assigning a unique IP address to the MANET interface, and a
prefix to the router); THEN we can configure HOSTS attached to this
router e.g. by using existing protocols like IPv6 SA". Even if the
difference is stated in section 5.1 of the manetarch draft, it could
be useful to remind the difference of autoconfiguration of the ingress
interface (thus the MANET interface) and the egress interface (thus
the attached network to the router -- which may be any kind of
network).

Regards,
Ulrich


_______________________________________________
Autoconf mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/autoconf