Re: [multipathtcp] A proposal for MPTCP Robust session Establishment (MPTCP RobE)

[Olivier Bonaventure <[e-mail]>]

Jing,
> 
> I meet the same problem that the connection establishment fails due to weak WiFi signal although LTE is good.
> The HappyEyeBalls is a good solution that we do not need modify the MPTCP. However, it requires the enhancement of the applications, as you mentioned.

On smartphones, most applications use a library that manages the 
underlying socket. By changing this library, you can enable most 
applications to benefit from the change.

> How about if the application does not support the enhancement, since we can't control the design of applications?
> I am considering that we may modify the MPTCP, e.g., let the MPTCP stack open two separate connections with indicated network IP addresses instead of the applications do. Moreover, this solution does not increase as much complexity as 'Duplicate SYN' does. Instead, the duplicate SYNs are sent to build two separate MPTCP connections without modifying the process of MPTCP connection establishment. How do you think?

It should be possible to change the implementation of the connect system 
call so that instead of selecting one source IP address and trying to 
create one connection it tries to create two MPTCP connections, 
typically one immediately with the default interface and the other one 
later after some timeout and then returns the first established 
connection and discards the other one (i.e. sends a RST upon reception 
of the SYN+ACK for the second).



Olivier