Re: [rtcweb] FW: I-D Action: draft-hutton-rtcweb-nat-firewall-considerations-00.txt

["Reinaldo Penno (repenno)" <[e-mail]>]

On 3/11/13 1:04 PM, "Harald Alvestrand" <[e-mail]> wrote:

>On 03/11/2013 08:52 PM, Reinaldo Penno (repenno) wrote:
>>
>> On 3/11/13 12:34 PM, "Hannes Tschofenig" <[e-mail]>
>>wrote:
>>
>>> On Mar 11, 2013, at 2:02 PM, Reinaldo Penno (repenno) wrote:
>>>
>>>>>> Why not use Port Control Protocol (PCP) to control Firewalls and
>>>>>>NATs
>>>>>> explicitly?
>>>>> We can switch to that as soon as 100% of firewalls support it - until
>>>>> then, we have to be able to rely on other techniques.
>>>> I'm sure STUN and TURN servers are not universally deployed ('100%')
>>>>in
>>>> ISP networks either.
>>> STUN and TURN don't require any support from ISPs.
>> If ISPs want to provide RTCweb like services don't they need STUN and
>>TURN
>> Servers so that ICE can gather candidates?
>
>ISPs may want to offer services. But that's independent of their role as
>ISPs.
>Anyone can offer a STUN or TURN server, and they can be anywhere on the
>Internet.
>
>PCP, on the other hand, has to be available on the specific firewall or
>NAT box you intend to traverse. If it isn't there, it won't work.

Yes, but on the other hand you deterministically get an IP address:port or
pinhole (both for incoming and outgoing connections) for a specific
lifetime instead of relying on outbound connections, keep-alives, and
external server. 

The point is that if the FW/NAT supports PCP, the solution is certainly
cleaner. Client is always free to fallback to STUN/TURN/indirect ways.


>>> Both protocols are used today.
>> Yes, today. But that did not stop design decisions to include these
>> protocols in ICE at a they time were not deployed at all.
>
>Sorry, I can't parse that.
>
>ICE was deployed to support applications that needed ICE; there was no
>need to deploy more than 1 STUN/TURN server in order to start using STUN
>and TURN.
>