Re: [rtcweb] Retry: DTLS carrying RTP/SAVPF over ICE: To UDP or not to UDP?

["Karl Stahl" <[e-mail]>]

Below I am agreeing with Harald that WebRTC browsers need to auto-discover
and use network provided TURN servers in favor of application provided TURN
servers (that will still remain with its hassles and limitations, for some
time of course). -->

/Karl

-----Ursprungligt meddelande-----
Från: Harald Alvestrand [mailto:[e-mail]] 
Skickat: den 12 juni 2014 16:58
Till: Pal Martinsen (palmarti); Karl Stahl
Kopia: Makaraju, Maridi Raju (Raju); Sergio Garcia Murillo; Javier Cerviño;
[e-mail]
Ämne: Re: [rtcweb] Retry: DTLS carrying RTP/SAVPF over ICE: To UDP or not to
UDP?

On 06/12/2014 01:30 PM, Pal Martinsen (palmarti) wrote:
> On 12 Jun 2014, at 11:58, Karl Stahl <[e-mail]> wrote:
>
>>>>     F18     The browser must be able to send streams and
>>>>             data to a peer in the presence of NATs and
>>>>             Firewalls that block UDP traffic.
>>> Yes. Using TURN servers is the way in which it is possible to fulfil 
>>> this
>> requirement.
>>
>> Further in draft-ietf-rtcweb-use-cases-and-requirements-14:
>> "the straddling TURN server ... It must be
>>    possible to configure the browsers used in the enterprise with
>>    network specific STUN and TURN servers.  This should be possible to
>>    achieve by auto-configuration methods."
>> F20     The browser must support the use of STUN and TURN
>>            servers that are supplied by entities other than
>>            the web application (i.e. the network provider).
>>
>> This is a TURN server with one interface on the Enterprise LAN where 
>> the browser simple shall place the WebRTC media. (The other interface 
>> should be public.)
>>
>> The way for the browser to autodiscover such network provided (from 
>> the enterprise and ISP) TURN server is being worked in 
>> draft-patil-tram-turn-serv-disc-01.txt.
>>
>> This resolves any firewall restriction not allowing WebRTC media (by 
>> paralleling the firewall with the TURN server), and also allows the 
>> enterprise or ISP to point out a better path for the WebRTC media 
>> (instead of through the often data crowded firewall default gateway).
>> (This will also be needed with IPv6.)
>>
> I have problems understanding how this solves any issues. I doubt many
enterprise network managers would install a TURN server in parallel with
their FW. Would be fair easier and safer to open up the appropriate ports in
the firewall towards the TURN server.  They could then allow only UDP to
allow for DPI and see that it is actually RTP traffic flowing.

That is actually a very convenient way to (logically) install a TURN server
in parallel with your firewall.
They don't have to be on the same physical segments to be in parallel.

The important thing that Cullen (I think it was Cullen) pointed out when we
discussed this requirement is that the TURN server is designated by the
local network administrator, not by the entity responsible for the Web
application that uses WebRTC.

[Karl] I fully agree with this - The WebRTC application should be off-loaded
of having to provide a TURN-server to allow usage behind (some)
NAT/firewalls (and we should not put that burden on the peer either). This
is something we do today to get something going (but still don't succeed
with the most restrictive firewalls), but hopefully don't have to do in the
future on "WebRTC-ready" accesses. Such accesses simply put ICE initiated
real-time traffic through the TURN server, while data goes as usual through
the default gateway (in a future clever firewall/NAT this can be the same
default gateway IP-address).

[Karl] Yes, let's hope for network provided turn servers, which could be
boxes in parallel with existing firewalls or built into the firewall. (A
turn server is much cleaner and better standardized to integrate, than
non-working SIP ALGs, where we instead put SBCs in parallel with firewalls
just to get SIP voice through (and with quality)). 

[Karl] Also consider all triple play (surf/telephony/IP-TV) broadband
accesses, that have a quality pipe just to cope with voice. A turn server in
such CPE will be required to place demanding WebRTC media on the pipe that
can handle WebRTCs real-time traffic. Separate pipes are also often deployed
e.g. for SIP trunking enterprise voice (PBXs and UC) paralleling the
firewall. A local turn server could similarly feed WebRTC media into such
pipes, then bypassing the ordinary firewall (that may be too restrictive to
allow real-time traffic or just is too data crowded to cope with WebRTC
media - NAT/Firewalls are typical congestion points).

Harald> Thus, some way for the browser to find this TURN server is needed.

[Karl] Auto discovered, network provided (enterprise or ISP) turn servers
are being done in draft-patil-tram-turn-serv-disc-01.txt. I favor the
anycast method suggested. Will WebRTC browsers implement all auto discovery
methods suggested there?

[Karl] Network provided turn servers available for the network users, need
no authentication - The WebRTC browser can just use them as part of the
Internet access (you get an IP address, a default gateway and a turn server
address to simply use). Those turn servers are required/can solve both
traversal and quality problems of many internet accesses.


>
> Having a TURN server supporting UDP/TCP/TLS as communication between the
client and TURN server would solve almost all connectivity problems. We
should, as Harald also pointed out, not try to force us throughout the FW if
the admin do not want the traffic to go through.
>
> .-.
> Pål-Erik
>
>
>
>
>
>
>
>
>
>
>> /Karl
>>
>> -----Ursprungligt meddelande-----
>> Från: rtcweb [mailto:[e-mail]] För Harald Alvestrand
>> Skickat: den 12 juni 2014 06:17
>> Till: Makaraju, Maridi Raju (Raju); Sergio Garcia Murillo; Javier 
>> Cerviño
>> Kopia: [e-mail]
>> Ämne: Re: [rtcweb] Retry: DTLS carrying RTP/SAVPF over ICE: To UDP or 
>> not to UDP?
>>
>> On 06/11/2014 08:38 PM, Makaraju, Maridi Raju (Raju) wrote:
>>>> The MUST was a consensus call in London. See the minutes for 
>>>> numbers. I
>> was one of >the ones who were surprised at the strength of the consensus.
>>> [Raju] I thought one of the implicit (may be explicit in
>> draft-ietf-rtcweb-use-cases-and-requirements-14) requirement for 
>> WebRTC is "implementations should work right out of the box without extra
boxes (e.g.
>> TURN) in between under UDP restrictive firewalls".
>> Unfortunately, nobody's found a way to make that true even for all 
>> common NATs/firewalls that want to permit the communication.
>>
>> TURN servers are the least obnoxious way to make sure apps can work 
>> in the presence of symmetric NATs.
>> We could wish that this wasn't so, but we could also wish that NATs 
>> would go away and everyone would run IPv6; neither is likely to happen
this year.
>>
>> Of course, in the presence of firewalls where the admin does *not* 
>> want to permit the communication, we neither can nor should make
communication work.
>>
>>
>>> Then I see the following requirement in
>> draft-ietf-rtcweb-use-cases-and-requirements-14. It did not mention 
>> use of TURN in the text.
>>>     F18     The browser must be able to send streams and
>>>             data to a peer in the presence of NATs and
>>>             Firewalls that block UDP traffic.
>> Yes. Using TURN servers is the way in which it is possible to fulfil 
>> this requirement.
>>
>>> BR
>>> Raju
>>>
>> _______________________________________________
>> rtcweb mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/rtcweb
>>
>> _______________________________________________
>> rtcweb mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/rtcweb