Re: [multipathtcp] charter discussion

[Olivier Bonaventure <[e-mail]>]

Behcet,
>
>
>         Regarding the charter discussions that happened at the mptcp
>         session on
>         Thursday March 30, I asked a question but I could not understand the
>         answer, let me ask it again.
>
>         In chairs slides page 12, TCP SYN shown in Option 1 and Option 2
>         actually were different although the slides seems to show that
>         they are
>         the same.
>
>
>     Option 2 is SOCKS. Option 1 places control information in the SYN
>     and SYN+AC while SOCKS places this information in the bytestream.
>
>         In Option 1, if the proxy sends TCP SYN than is it OK in a
>         conversation
>         the destination does not know who the source is?
>
>
>     This is common with proxies. SOCKS and HTTP proxies for example
>     often use their own address to contact
>     remote servers and not the client address. draft-plain-mode proposes
>     a method to allow the proxy to learn the client address and thus use
>     it to reach the final detaintion
>
>         In Option 2, if TCP SYN sent was the original TCP SYN from the
>         source then
>         how come the TCP-ACK goes through the proxy?
>
>
>     Option 2 is SOCKS, thus the proxy uses its own address as a source
>     to contact the final server.
>
>
> So both Option 1 and Option 2 sends out TCP SYN with proxy own address
> as the source. I thought the two TCP SYN s were different?

In Option1, this depends whether the source address has been included in 
the SYN as explained in draft-plain-mode

>
>
>
>         A node sending a packet without its address as source address
>         means it
>         is router, right?
>
>
>     The node could have a pool of addresses like a CG-NAT
>
>
> The way I understand is TCP SYN from each user (possibly many per
> user) is effectively tunneled to the proxy using different mechanisms in
> Option 1 and Option 2, actually Option 2 makes it more implicit.
> It seems like not only TCP SYN but all three-way TCP handshake needs to
> be tunneled.

There are no tunnels.

>
> CPE has to deal with many parallel MPTCP connections to the same
> endpoint, the proxy. I think that HTTP proxy does not have this type of
> problem.

The HTTP proxy only carries HTTP traffic. The proxy described in 
draft-plain-mode works with anytype of TCP application.


Olivier