Re: [Autoconf] tentative draft-ietf-autoconf-problem-statement-03.txt and the manetarch document

[Shubhranshu <[e-mail]>]

Alex,

Please see below.

----- Original Message -----
From: "Alexandru Petrescu" <[e-mail]>

> Shubhranshu wrote:
> [...]
>> I can see the problem being described here but yes I agree that it
>> needs to be expanded, to further calrify things. See below.
>
> Shubhranshu, thank you for reading my proposal text.  I'm taking the
> comments.  I think I will make modifications as you suggest (see below)
> but I don't think I will do them right away.  I think I will do that for
> the next next IETF (after Vancouver), to let time for things to settle.

Thanks. I sent some of my thoughts a while before hoping that it would
help continue our discussion and make some progress even before the
Vancouver meeting. I look forward to hear your comments as well....:-)

>
> Besides, there may be some impact in how the manetarch document is
> written: should the manetarch document picture a non-hierarchical
> addressing architecture (in addition to the pictured _hierarchical_
> addressing architecture Figure 6).  I think yes because that (manetarch)
> draft says that "MANETs of those sizes can perform reasonably well in
> many cases w/o hierarchy" - what does "hierachy" mean?

I think you are talking about section 8.2. Its Routing hierarchy.

<snip>
> Shubhranshu, I can try to re-formulate that text to express better that
> that is a problem not a solution.  If one tries to run DHCPv6 Server and
> a DHCPv6 Relay, and then a Router behind the Relay requests for a prefix
> (with DHCPv6 PRefix Delegation) then one realizes that (1) the allocated
> prefix must be hierarchically[*] fit within the prefix administratively
> assigned on the Relay's interface towards Router and (2) once the prefix
> is allocated it will unfortunately _not_ be inserted in DHCP Server's
> routing table (because it is already covered by the routing entry
> pointing to Relay's prefix).
>
> This is a problem from implementation.
>
> It could help if we had a requirement (within a bulleted list of Req
> items) on architecture of unicast prefixes assigned on MANET subnets to
> be able to be both hierarchical and non-hierarchical - i.e. there should
> be no strict hierarchical relationship between IPv6 prefixes assigned to
> the MANET subnets (even when the pictured topology seems hierarchical
> tree-like).

Above you said "the allocated prefix must be hierarchically[*] fit.."
but your requirement in the immediately above paragraph is "there
should be no strict hierarchical relationship between IPv6 prefixes
assigned". I guess I am missing something since these two sounds
contradicting to me.

>
> To better explain that need to support non-hierarchical addressing
> architectures one would first have to explain hierarchical address
> architecture and then, keeping the same topology, one would do a
> non-hierarchical addressing architecture.  I describe this below.
> (Remark that the manetarch draft only pictures a _hierarchical_
> addressing architecture in Figure 6.)
>
> An example of a non-hierarchical address architecture followed by a
> hierarchical address architecture is illustrated below.  Assume three
> prefixes: 8000::/length (also represented as "100..." meaning first 3
> bits are "100" and then followed by 125 0s), 4000::/length (010...) and
> c000::/length (110...) and three MANET Routers.
>
> Hierarchical IPv6 address architecture on a topology illustrated as a tree:
>                            ------
>                           |MANET |
>                           |Router|
>                            ------
>                          /        \
>               8000::/1  /          \ 4000::/1
>               (100...) /            \(010...)
>                     ------        ------
>                    |MANET |      |MANET |
>                    |Router|      |Router|
>                     ------        ------
>                        |
>                        | c000::/2
>                          (110...)
>
> Same topology illustrated linearly, and with a IPv6 non-hierarchical
> addressing architecture:
>
>                 ------            ------            ------
>         -------|MANET |----------|MANET |----------|MANET |
>        8000::/1|Router| 4000::/1 |Router| c000::/1 |Router|
>        (100...) ------  (010...)  ------  (110...)  ------
>
> Both types of IPv6 addressing architecture should be supported by an
> address auto-configuration protocol for MANETs.  This is a requirement,
> within a buletted list.

Thanks. You explained very well but I am not yet decided on this.
Perhaps I need to further analyse both of them. I'll do that.

- Shubhranshu

>
> (if/when this description of hierarchy is agreed then I can suggest
>  another useful addressing architecture from MANEMO).
>
> For solution, for example, one could write spec to say that a MANET
> Router acting as a DHCP Server and requesting PD requests from a MANET
> Router acting as a Relay will always update its forwarding information
> base with respect to that Relay and the allocated prefix.  This is
> indeed part of solution.
>
>>>>    These are two issues that appear in the topology pictured in Figure
>>>>    1.  In this figure the DHCP Server, Border Router, MR1 and MR3 are
>>>>    assumed to be stable non-moving (only MR2 moves).  If we consider
>>>>    that any entities will also move then there may be more DHCP issues
>>>>    that could be identified.
>>
>> Yes.
>
> Ok.  I think some more could be identified.  Emmanuel asked the same
> thing.  I think it will take some time to identify them, that's why I
> postpone it to next next IETF.
> Please accept my excuses if this looks too pedagogical... it's just my
> means to communicate...
>
> Alex


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