Re: [babel] Robert Wilton's Discuss on draft-ietf-babel-rtt-extension-05: (with DISCUSS)

[Juliusz Chroboczek <[e-mail]>]

> 1) the bits required for interop (wire format and required handling)
> 2) the algorithm suggested to use the bits from (1)

> Conceptually, (1) is standards track and (2) is experimental.

Part 1 defines a packet format and an efficient algorithm to compute RTTs.
I agree with David (and the WG) that it should be standards track, so that
people can build on it.

Part 2 describes a specific algorithm that uses RTT for route selection
and that happens to work well in practice.  Other algorithms for route
selection could be designed without impairing interoperability (Babel does
not require that all routers in a routing domain use the same route
selection algorithm), and RTT could be used for other applications
(e.g. for monitoring).

I do not think that part 2 needs to be normative, since it is not required
in order to make implementations interoperable.

> I'll note that (2) here is in better state than many other algorithms
> I've seen in other standards track documents; the warning disclaimer
> mainly comes from the fact that one of the authors likes his
> mathematical proofs and isn't fully satisfied with an algorithm that
> only happens to work really well in real deployments, instead of that
> plus a mathematical proof.

Yes, the algorithm has been deployed on thousands of routers since 2014,
with no issues known to us.  We've also been unable to break it in the
lab, no matter how hard we tried.

As David says, we don't have a theoretical understanding of why it works
as well as it does.  Together with the point about interoperability made
above, this implies to me that part 2 should not be made normative.

An obvious solution would be to write two RFCs, one normative and one
experimental, but the two parts really belong together, so I'd much prefer
them to remain in one document.  The expedient solution would be to
rewrite part 2 in normative language, but that would not be entirely honest.

-- Juliusz