Re: [multipathtcp] Consensus call on potential MPTCP proxy work

[Olivier Bonaventure <[e-mail]>]

Costin,

>> I fully support the work on MPTCP proxies. I would not focus on the current use case (how gateway and two proxies) but more on the importance of :
>>
>> - being able to terminate Multipath TCP connections on a proxy somewhere in the network to bond different access networks when the server is not MPTCP capable
>> - using 0-rtt, i.e. the proposed solution should not require an additional rtt to create a connection
>
> + 1.
>
> Also, the solution should be future proof - if the server does support MPTCP, the proxy should simply relay the traffic, and not terminate the connection, allowing the client to directly connect to the server with other subflows.

Since the proxy is explict, the client creates an MPTCP connection to 
the proxy and the proxy creates another MPTCP connection to the final 
server. Since the proxy was the destination of the SYN, it cannot simply 
relay the traffic, but it should indicate to the client that the server 
supports MPTCP so that the next connection established by the client 
towards this server can directly bypass the proxy. This information 
about the MPTCP support on the final server could be part of the 
signalling information contained in the payload of the SYN+ACK



Olivier