Re: [Idr] RS BFD draft

On Fri, Mar 31, 2017 at 09:34:08AM -0500, Robert Raszuk wrote:
> To your point of assymetry ...
> 
> If the assymetry happens between clients placed in different rib groups of
> the given RS to corellate it is really ugly and breaks all today's code
> running as BGP route servers in the field.
> 
> Besides for say UDP streaming unidirectional connectivity is not that bad
> to withdraw reachability just because symmetry is not there.
> 
> Last most IX clients may have other connectivty outside of IX fabric in one
> or both directions so even TCP would work.

For posterity for the mailing list, Robert and I had this discussion
off-list.

In particular, this refers to the point covered in section 4.2 of the RS-BFD
draft.  This is where the route server may tell one peer about routes with a
given nexthop but not tell the other peer about *any* routes with the
nexthop for that first peer.  Since BFD requires a pair of endpoints over
which to provision a session, it is necessary for the route server to make
sure both sides are informed about both sides/nexthops even if routing only
advertises them one way.

> Bottom line I agree on the need to document as BCP NH tracking part of the
> draft between IX clients, but new SAFI seems to me not to be a best idea.
> 
> Best,
> R.
> 
> PS. If client does not support add-path it can use different sessions as
> described by RFC6774.

The RS-BFD proposal doesn't preclude other mechanisms for advertising
alternate paths.  I'd be interested in hearing about RS operators that have
any intentions of deploying BGP add-paths[1] or diverse paths.

-- Jeff

[1] It should be noted that add-paths is considered problematic in the
traditional eBGP route distribution scenario.
draft-pmohapat-idr-fast-conn-restore was originally drafted to address that
case for traditional service provider networks.  

> _______________________________________________
> Idr mailing list
> Idr@ietf.org
> https://www.ietf.org/mailman/listinfo/idr