Re: [Idr] Capability Advertisement in draft-ietf-idr-bgp-extended-messages

[Jeffrey Haas <[e-mail]>]

Enke,

On Wed, Jul 31, 2019 at 09:50:08PM +0000, Enke Chen (enkechen) wrote:
> >>  Note that RFC 6793 (4-byte ASes) require bi-directional advertisement.
> 
> No, this statement is not correct. It is fundamental (in transition) for a BGP  speaker
> to be able to talk to both NEW speakers (that have advertised the capability), and OLD
> speakers (that have not advertised the capability).  Different encodings are used in the
> UPDATE message depending on whether the 4-byte AS capability is received from a
> neighbor.

I should have known I wasn't pedantic enough in this comment. :-)

The point here is that in order to exercise the procedures between NEW BGP
speakers, (RFC 6793, §4.1), both sides must advertise and use the
capability.  If you have a mix, each speaks 4271 to each other with the new
speaker running the transitional procedures.

With regard to the extended messaging, my preference is that both sides
advertise the capability in order to use the large messages.  A mis-match
falling back to 4271 4k PDUs is fine - symmetrically.  Asymmetrically
sending extended messages leads to a mess of edge cases.

-- Jeff