Re: Re: [Fwd: New Liaison Statement, "LS on need for Multi-Path QU IC for ATSSS"]

[Olivier Bonaventure <[e-mail]>]

Martin,

> On 2 Apr 2020, at 01:56, Martin Thomson <[e-mail]> wrote:
> 
> If I read this liaison correctly, this is effectively a request to keep 3GPP in the loop.  The request to do multipath is more implied than anything else.  I think that the explicit request is fine.  The implicit one maybe less so.
> 
> I suspect that work here would go faster if we had more input on the hard multipath problems.  That's not the spelling of the protocol (that's almost trivial, as demonstrated in proposals), but the assignment of different information elements paths that might have different characteristics.  What I heard in Zürich was that this was less of an engineering problem than a research one.
> 
> I'd be interested in having active engagement from people who have use cases. Specifically, I would seek to learn whether this is truly a research question or whether there are simplifying assumptions that might make the problem tractable for some use cases.

Since the publication of Multipath TCP in RFC6824, we have accumulated experience in deploying and operating solutions that leverage multipath transport. Here are a few pointers to documents that describe these deployments. Several of the people that were involved in multipath TCP and also active in QUIC and can bring their experience to finalise multipath extensions to QUIC.

https://www.ietfjournal.org/multipath-tcp-deployments/ <https://www.ietfjournal.org/multipath-tcp-deployments/>
https://tools.ietf.org/html/rfc8041 <https://tools.ietf.org/html/rfc8041>
https://arxiv.org/abs/1907.04570 <https://arxiv.org/abs/1907.04570>

I can provide additional information for some of these use cases

> The more I think on this subject, the more that I think it was a mistake to charter to do multipath.  I don't believe that we properly answered the standard BoF questions with respect to this topic.

I disagree. Given the flexibility of QUIC, it would be a mistake to not consider multipath. TCP should have been a multipath protocol from day one.


Olivier



-- 


Disclaimer: https://www.tessares.net/mail-disclaimer/ 
<https://www.tessares.net/mail-disclaimer/>