[Anima] draft-richardson-anima-l2-friendly-acp-02.txt

[Michael Richardson <[e-mail]>]

WG, after some thinking, I have revised this document from problem statement
to a proposed solution.

The short of it is that the problem is multicast/broadcast, such as done for
ND, DAD and GRASP-DULL.   Unicast traffic between two control plane CPUs
should not be a problem, so actual ACP traffic could be sent just fine, as
long as it is only unicast.

So the problem isn't ACP operation (IKEv2, IPsec, TCP for BRSKI onboarding),
but GRASP-DULL discovery!   This was my thought a few weeks ago.

My answer is to embed the GRASP DULL CBOR inside an LLDP TLV.
The only only issue is that each GRASP M_FLOOD message contains a session-id,
which I think needs to change upon each M_FLOOD that is sent.  It would be
better for LLDP, as I understand it, if each message could be identical,
unless there is an actual change to the contents.
(Annoyingly, "session-id" does not appear in RFC8990 Section 2.7 on Session
Identifier)

I can't identify an impact of not changing it when the LLDP is retransmitted.


[e-mail] wrote:
    > Title:		Autonomic Control Plane design for Layer-Two Switched Networks
    > Document date:	2021-06-15
    > Group:		Individual Submission
    > Pages:		6

(it got shorter!)

    > Html:           https://www.ietf.org/archive/id/draft-richardson-anima-l2-friendly-acp-02.html


    > Abstract:
    > This document proposes a design for an L2 aware Autonomic Control
    > Plane that can be deployed easily to layer-two (Ethernet) switched
    > technologies that are common on Campus/Enterprise network
    > architectures.

    > This document leverages the hop-by-hop announcement used in LLDP, but
    > runs bulk data over normal IPv6 Link-Local unicast ethernet frames.






--
Michael Richardson <[e-mail]>   . o O ( IPv6 IøT consulting )
           Sandelman Software Works Inc, Ottawa and Worldwide