Re: [multipathtcp] draft-boucadair-mptcp-plain-mode-07

[Joe Touch <[e-mail]>]

On 7/1/2016 12:01 AM, Olivier Bonaventure wrote:
> Hello,
>>
>> In the paired MPTCP proxies mechanism, there is user's TCP session
>> and there is also multipath TCP session. It is a TCP over TCP idea,
>> correct?
>
> No, there are three TCP connections in this scenario :
>
> - one (regular) TCP connection between client and CPE
> - one Multipath TCP connection between the CPE and the Concentrator
> - one (regular) TCP connection between the Concentrator and the remote
> server
>
>> As for TCP over TCP, the following page explains why an internal
>> meltdown effect would take place due to the interaction between two
>> leveled retransmission mechanisms.
>> http://sites.inka.de/bigred/devel/tcp-tcp.html
>
> This is a well known issue.
>
>> Do we have solid solutions for this issue nowadays?
>
> The solution is to terminate the TCP connections transparently. This
> is what most TCP optimisers used in mobile networks do already.

Is there a current standard or BCP endorsing this practice?

AFAICT, it's in direct opposition of RFC1122. I hope we don't want to
create a new RFC that flies in the face of what few standards we have
unless we're prepared to revise those standards first.

"deployed in the wild" is not a sufficient justification, IMO.

Joe