Re: [babel] Do we want to adopt draft-jonglez-babel-rtt-extension?

[Toke Høiland-Jørgensen <[e-mail]>]

Juliusz Chroboczek <[e-mail]> writes:

> Dear all,
>
> Of all the Babel protocol extensions, the ones that are used in production
> are all IETF WG documents.  Except draft-jonglez-babel-rtt-extension.
>
>   https://tools.ietf.org/html/draft-jonglez-babel-rtt-extension
>
> Draft-jonglez-babel-rtt-extension consists of one non-mandatory sub-TLV
> that can be carried on Hello and IHU packets.  It contains enough data to
> apply Mills' algorithm (the one used in the HELLO routing protocol and
> later in NTP) in order to compute an estimate of RTT with microsecond
> granularity.  The draft only specifies how to compute RTT -- using RTT for
> something useful is left to the implementation -- but it gives an example
> of using RTT in order to derive link cost.
>
> The algorithm is described in more detail in
>
>     https://arxiv.org/pdf/1403.3488.pdf
>
> The actual motivation is overlay networks, where a single hop might span
> continents, and least-hop routing yields suboptimal results.  The
> algorithm has been used in production for a few years now by a French
> company called Nexedi.
>
> The same subprotocol could probably be used in other kinds of networks,
> most notably congested wireless networks.  More experimentation is needed.
>
> Personal opinion -- the algorithm as described in the draft has one flaw:
> since the IHU sub-TLV does not carry a Transmit timestamp, the IHU sub-TLV
> is only useful in cases where the same Packet contains a Hello TLV.  I'm
> not sure if people would be happy with a flag day, so fixing that would
> require either eating up another sub-TLV number, or putting the Transmit
> timestamp at the very end of the sub-TLV (which would make for an ugly
> format).
>
> Do we want to adopt this draft?  I can easily upload a new revision before
> Monday's cutoff date.

I was planning to eventually implement this anyway, so sure, I support
adoption :)

-Toke