On Sun, Sep 27, 2009 at 1:13 PM, jc <[email protected]> wrote:
>
> On Sep 27, 2009, at 12:26 PM, Eric Rescorla wrote:
>
> Every time you perform an Attach in RELOAD it uses either DTLS or TLS  with
> STUN to perform ICE connectivity checks(done by TLS/DTLS hand-hake in
> ice-lite) towards the remote candidates. Is this fact or fiction?
>
> It's fact but you're confusing two meanings of the word "with". In this case
> tls/dtls is multiplexed with stun over tcp/udp, therefore there is no need
> to define stun over dtls. The stun over tls you are pointing to is not used
> with ice, but rather for other stun-using applications such as turn
>
>
> I've built out my transports so that the lowest layer is UDP/TCP the next is
> DTLS/TLS and the next demultiplexes STUN, RELOAD and APP. In other words
> DTLS and TLS are completely transparent to the ICE transport. I don't
> demultiplex any TLS or DTLS packets because when the SSL state is acquired
> everything else can only be upper layer messages. What you've stated
> indicates to me that you may receive a STUN message on the UDP/TCP layer
> because you've mentioned having to "multiplex tls/dtls" with stun over
> tcp/udp". Does my implementation sound correct to you?

No, this doesn't sound right.

The idea here is that ICE provides a fast connectivity check and then
you only establish TLS/DTLS connections over the single channel
that ICE successfully establishes. If you do things the other way
(TLS first, then ICE), then you end up trying to establish a zillion
independent (D)TLS connections (which would be really slow even
if it worked, which it won't) and then running ICE, which is pointless
since you can't establish a (D)TLS connection unless you have two-way
connectivity.

-Ekr
_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to