[mat-wg] followup question from MAT-WG presentation on

Hello Benno,
Thank you for the question and thanks to Brian for forwarding it.

  I would say UDP-O have chances of being deployed because they currently 
work on a majority of paths (at least using CCO).

  Other than the paths tested in our work, they also work for example in 
mobile networks (obviously not all of them) where it is very hard that 
transports other than TCP or UDP work.

  The MPTCP approach you pointed is very interesting and I would say the 
similar thing for UDP-O is that UDP-O is still UDP (like MPTCP is still 
TCP) so it can traverse quite easily except for the cases in which there 
is a length consistency check.

  I think mostly depends on the applications using UDP-O. If there will 
be some strong use cases for UDP-O (for example DNSSEC) we could 
probably see UDP-O deployed.

Cheers,
Raffaele Zullo



On 2020-10-28 16:26, Benno Overeinder wrote:
>> On 28 Oct 2020, at 17:02, Brian Trammell (RIPE) 
>> <ietf+ripe at trammell.ch> wrote:
>> 
>> hi Raffaele,
>> 
>> Due to time restrictions, we missed the following question from your 
>> MAT WG presentation today, for discussion on the list:
>> 
>> "Do you think UDP-O will be successfully deployed in the future? Think 
>> of the sctp protocol and how multipath-tcp learned from the deployment 
>> barriers."
>> 
> 
> Also considering the failure in the adoption of the SCTP protocol by
> middleboxes dropping traffic with a new (unknown to them) transport.
> Multipath TCP learned from this, and in the Multipath TCP
> architecture, the design team retrofitted the multiple streams into
> independent TCP streams, just like regular TCP.
> 
> See also the overview paper by Christoph Paasch and Olivier
> Bonaventure, Multipath TCP: Decoupled from IP, TCP is at last able to
> support multihomed hosts, ACM Queue, vol. 12, no. 2, March 2014.
> (https://queue.acm.org/detail.cfm?id=2591369)
> 
> Cheers,
> 
> — Benno