[Serusers] mediaproxy feature request: immediate re-invite
Klaus Darilion
klaus.mailinglists at pernau.at
Tue Sep 6 12:25:03 CEST 2005
But how do you know if a client is behind symmetric NAT or a less
restrictive NAT - that means how will you know if the reINVITE with
changed SDP will work?
klaus
Federico Giannici wrote:
> Klaus Darilion wrote:
>
>> If you have symmetric NAT, then this wont work (as you already said).
>>
>> If you have less restictive NAT, you can use STUN so there is no need
>> for a mediaproxy at all.
>
>
> OK, but I'd like to handle the NAT problems on the server side instead
> of the clients, because:
>
> 1) It simplifies clients configuration.
>
> 2) I have found that many STUN client implementations are bugged and
> even a GREAT number of ALG implementations are bugged too.
>
> 3) I'd like to solve Hairpin problems too, as it is one of the biggest
> problem I have encountered.
>
> With regard of the last point, with the nathelper module I can solve the
> problem of not knowing the NATted SIP port and with this feature of
> mediaproxy I could solve the problem of not knowing the NATted RTP port.
>
> So, making the clients register with their private (not NATted) IP, I
> could have all the data for normal working, and moreover, having the
> original (not NATted IP/ports), I could also avoid the hairpin problems
> (at least for users behind only one level of NAT).
>
> Bye.
>
>
>> Federico Giannici wrote:
>>
>>> As I found no other support email address for mediaproxy, I'm sending
>>> here this request...
>>>
>>> Is it possible to implement an option in mediaproxy to allow an
>>> IMMEDIATE RE-INVITE to both UAs as soon as the RTP streams start?
>>>
>>> The idea is to use Mediaproxy only to find out the NATted RTP ports
>>> of both UAs. Then a re-INVITE is immediately sent to the UAs so that
>>> the streams are no longer proxied, avoiding all problems of RTP
>>> relaying: waste of bandwidth, longer delays, bad scalability, etc.
>>>
>>> I know this will not work with symmetric NATs, but I found that they
>>> are relatively rare and anyway they will neither work with STUN...
>>>
>>>
>>> If anybody knows that this is theoretically impossible to implement,
>>> please tell me that.
>>>
>>>
>>> Thanks.
>>>
>>
>>
>
>
More information about the sr-users
mailing list