Re: updated TURN draft

	Hello again,

I don't understand how TURN clients could receive incoming TCP
connections with the new draft 08. Maybe I'm getting it wrong.

According to the draft, if a TCP connection is received by the TURN
server before the TURN client made a Send Request, the connection is
rejected. However, if the TURN client makes a Send Request before the
TURN server receives the TCP connection, then the TURN server will try
to establish the connection actively, instead of waiting for an incoming
connection from the specified host.

It sounds like a vicious circle. In particular, I wonder how two TURN
clients could establish a TCP connection between themselves with the new
scheme.

As a side note, I'm doubtful about the use of TURN server to relay
active TCP connections. Such connections should go through NATs without
the help of a TURN server, and as such, not using the TURN server would
save TURN server's bandwidth.

There doesn't even seem to be the advantage of knowing the external
source (IP, port) tuple given the TURN server doesn't tell the TURN
client about the used "eph[me]eral port" that it has used.

Regards,

--=20
R=C3=A9mi Denis-Courmont <[email protected]>
Nokia-NRC/Helsinki