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

[Juliusz Chroboczek <[e-mail]>]

> My initial conclusion is that this work is either (a) not needed at all,
> (b) out of scope, or (c) involves one if not two very bad ideas.

Very well put.  I'll add two more points.

First, the plain mode draft requires doing some fairly low-level things,
such as proxying a SYN and the corresponding SYN-ACK separately.  This is
not mere transport-layer proxying, and feels way too close to NAT for my
comfort.  Since in at least some deployments the plain mode proxy is
on-path, and cannot be easily circumvented by the client, it will cause
all the problems with middleboxes that this particular group is so sadly
acquainted with.

If we're going to do NAT for IPv6, then perhaps we shouldn't be doing it
in this group.  Or at all.

Second, I don't understand whether this work is specific to MPTCP.  As far
as I understand, the proposal is about interception proxying at the
transport layer, and does not rely on MPTCP: the link between the proxies
could very well be using SCTP, or MP-QUIC, or whatever the next fashionable
multipath protocol will be.  As such, I'm not sure it's in scope.

I'm opposed to this work, at least pending further clarification.

-- Juliusz