Re: [Anima] big pictures diagrams --- how bootstrap fits into signaling and discovery

[Brian E Carpenter <[e-mail]>]

Hi Michael,

Cherry-picking in your message:
On 06/07/2015 08:29, Michael Richardson wrote:
> 
> After reading Brians' slides last week, 

Those were draft slides sent only to the signaling design team.
When they're ready they will of course be for the whole WG.

> I felt that there was some bigger
> picture concepts/questions that were missing.  I drew these diagrams since.

Very interesting. This, IMHO, should be distilled into input to the reference
model draft.

> While I've discussed this with Max, they are my own views.
> 
> There are 9 of them, located at:
>   http://www.sandelman.ca/SSW/ietf/anima/diagrams/

...
>   I will also say that I think that discovery process should include running
>   DHCP(v6).  

Please explain how that works in a network with no administrator
and no pre-existing default configs. I don't see where a north-south
direction would even come from, let alone a designated DHCPv6
server.

> I'm also unconvinced that GDNP/GRASP shouldn't be other than
>   DHCPv6, 

We have a list of 25 requirements. I don't think DHCP fits them,
not even close. (I don't care about packet formats; in fact I'm
pretty sure we'll have use cases where carrying DHCP options in
anima signaling is entirely appropriate.)

> and if it should be other, then should be DNCP based.

I don't think DNCP fits the 25 requirements either, and I tried
quite hard to make it, since at least it isn't a north-south
protocol like DHCP. (Again, I don't care about packet formats;
the current GDNP TLV format is intentionally compatible with DNCP,
but Toerless has argued passionately for a change to JSON/CBOR.)

Please review the requirements and Appendix A in
draft-carpenter-anima-gdn-protocol-04, and if you can show in full
detail how DHCP(v6) or DNCP can meet the requirements, we can have
the discussion.

>   *DISCOVERY* should, I think, occur at layer 2 and layer 3, and the results
>   taken together.

Discovery of what? In Anima signaling, we are only interested in
discovering Anima signaling peers. We are not particularly
interested in topology or nearest-neighbor discovery. The routing
system may need that kind of discovery but IMHO that is simply
a different problem.
...
> 
> 4) http://www.sandelman.ca/SSW/ietf/anima/diagrams/network4-identity.svg
...
> Not shown is some process whereby the BGP peer
>    goes to talk to it's domain controller and learns that R2/R4 are not welcome.

I think that is the tricky bit - how does an entity know *in a zero-touch way*
which domain it is in and how to trust the domain registrar/controller?

>    Several people had expressed needs to have the ACP potentially traverse
>    domain boundaries, and this is an example of discovering a boundary.

I suspect that trust *across* domain boundaries will only be possible
with explicit Intent distribution (saying "it's OK to trust BGP speakers
in domain X presenting credenial Y").
> 
> 
> 5) http://www.sandelman.ca/SSW/ietf/anima/diagrams/network5-routing.svg
> 
>    On top of these point to point tunnels, an RPL (rfc6550) instance would
>    run, 

but... but... only some Anima networks will run RPL. We definitely need to
be agnostic about the choice of IGP. If we've learned one thing from
homenet, we've learned that, surely?

...

> 6) http://www.sandelman.ca/SSW/ietf/anima/diagrams/network6-imprint.svg
> 
>    Devices call "home".
>    In the first version of this diagram, the devices called the vendor A,
>    and vendor B boxes directly.  That could be the case, and they could be
>    redirected to the domain owner based upon sales records, or it could be
>    that the uplink, being provided by the domain owner, intercepts all
>    traffic and directs it to the registrar.  

"Intercepts" is a bad word. IMHO we need to devices to discover whether
there is a local registrar first, and not waste their time trying to
call home to a vendor in deployments where that is impossible. (There
is a chicken/egg problem here, since that should be a matter of Intent
but you can't get Intent until you have been authenticated.)

>    Or the domain registrar could
>    be running an ASA, and could have joined the ACP as well.

Definitely.

I will stop there because I think we need the ACP authors to
comment on much of what you said. But thanks for the systematic
thinking.

    Brian