Basically, I think that when you make a TCP connection,
the registrar keeps a note of it. So when a request comes
in for the registered user, the proxy just uses the connection
it has for that user.
By the way, in "outbound", these kept-alive connections are
called "flows".
>>Like UA makes register over TCP, what registrar does ? stores reference
>>to that TCP connection ?
Yes, the proxy stores the TCP, IP and Port of the registered user.
>> must it check live tcp connections first, if exists uses it.
yes.
>>If
>>doesn't exist
If the flow doesn't exist, then you can't make the call to the
registered UA and the proxy will have to reject the call.
However, the draft explains how UA's can detect if any flow
has failed. In a flow failure case, the UA will then re-register
as soon as it can.
>> (choose UDP if message small, try TCP for bigger message) ?
well, I think, for best results, always use TCP.
I think, for outbound, you would have to register a UDP flow and
a TCP one (if you wanted to use either) so it might just be simpler to
do just TCP.
Regards,
Attila
-----Original Message-----
From: Ivar [mailto:[EMAIL PROTECTED]
Sent: 20 April 2007 10:38
To: Attila Sipos
Cc: [email protected]
Subject: Re: [Sip-implementors] REGISTRATION TCP
This draft specifies keep-alive, but i'm confused about other thing.
Like UA makes register over TCP, what registrar does ? stores reference
to that TCP connection ?
Because most of times UAs behind nat, proxy can't connect UA with tcp,
UA must keep connection alive.
Something Like
Contact: TCP:IP:PORT will be stored with that conact ?
or how proxy chooses connection for that contact.
for example (AOR = sip:[EMAIL PROTECTED]) Conatct: sip:[EMAIL PROTECTED]
made over TCP -> connection alive,
now incoming INVITE to sip:[EMAIL PROTECTED], proxy will get Conatct:
sip:[EMAIL PROTECTED], now whats next ?
must it check live tcp connections first, if exists uses it. If
doesn't exist (choose UDP if message small, try TCP for bigger message) ?
I desperately trying to get right connection choosing plan.
Attila Sipos wrote:
> Have a look at the so-called "outbound" draft:
> http://www.ietf.org/internet-drafts/draft-ietf-sip-outbound-08.txt
>
>
> When you register with TCP, the TCP connection is kept-up with some
> kind of keepalive (see section 4.4. Detecting Flow Failure).
>
> With UDP, you can keepalive the connection with STUN
> or, though not recommended, you can keep the NAT-binding alive
> with frequent re-registration.
>
> Regards,
>
> Attila
>
>
> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] Behalf Of Ivar
> Sent: 20 April 2007 09:26
> To: [email protected]
> Subject: [Sip-implementors] REGISTRATION TCP
>
>
> Hi,
>
> Googling, seeing rfc 3261 seems i miss some thing ....
>
> How register method over TCP acts ?
> Does register contact binding must hold reference to TCP connection what
> made register ?
> If not then how then you never can connect LAN SIP device over TCP and
> UDP won't work too because phone haven't sent packet out.
>
> If registration keeps connection and all request to that SIP device will
> be sent through that connection i get how that will work.
> unknown UA 1 -> invite -> proxy, proxy will reuse registration TCP
> connection -> UA 2
>
> Any comments very welcome.
>
> _______________________________________________
> Sip-implementors mailing list
> [email protected]
> https://lists.cs.columbia.edu/cucslists/listinfo/sip-implementors
>
_______________________________________________
Sip-implementors mailing list
[email protected]
https://lists.cs.columbia.edu/cucslists/listinfo/sip-implementors
