Re: [dispatch] PCP for SIP Deployments

[<[e-mail]>]

Dear Cullen,

Thank you for the comment. 

I agree with you it is more robust to use PCP along with other mechanisms for the generic case. In such case, PCP offer other functionalities that are not listed in this I-D, e.g., NAT detect. 

I also agree that if the SIP UA does not use the PCP response to build its messages, then relying on rport and outbound will be needed even if PCP is used to instantiate the requested mappings. What we are suggesting in this draft is that we can disable those features if the SIP UA relies on the returned external IP Address, external port number, the pair(s) of ports that preserve parity to build an offer/answer.

We had this text in the I-D: 

   In deployments where ICE [RFC5245] is required, PCP can be of great
   help as discussed in [I-D.penno-rtcweb-pcp] for the WebRTC case.  ICE
   can be used in the context of SIP over WebSocket [RFC7118] and WebRTC
   when deployed within managed networks.  Because TURN suffers from
   limitations in traversing NAT and firewalls over UDP, PCP is a
   promising solution that can complement ICE in those deployment
   contexts to soften the experienced high failure rate [ICEFailure].

The point is that this draft was initially scoped to cover the very specific case of SIP service delivery in managed networks. In those networks, PCP proxies and/or IGD/PCP Inetworking functions can be enabled if needed and justified. E.g.,
* Deliver SIP service to a DS-Lite serviced customer: there is only one level of NAT that is in the network side. The CPE does not include any NAT. The SIP endpoint can be embedded in the CPE itself or be located behind.  
* Access to an IPv4 SIP service platform via a NAT64: Only one level of NAT will be experienced. This is more for the ongoing IPv6-only deployments in mobile networks. 

Thank you.

Cheers,
Med

> -----Message d'origine-----
> De : Cullen Jennings [mailto:[e-mail]]
> Envoyé : lundi 9 mars 2015 15:35
> À : BOUCADAIR Mohamed IMT/OLN
> Cc : [e-mail]
> Objet : Re: [dispatch] PCP for SIP Deployments
> 
> 
> I read the draft and it seems like one of the issues is that you don't
> know if the PCP nat is the only nat or firewall in path. So it seems like
> a more robust solutions is to use PCP along with existing solutions. So
> for SIP, one does the PCP to get a port but still relies on things like
> rport and outbound to correctly set the return address (IE use the PCP to
> open a pin whole but still use private address in contact). Similarly with
> the RTP use PCP to get a address on the outside of the NAT but just add
> that in as one of the ICE candidates.
> 
> 
> 
> 
> > On Feb 26, 2015, at 2:31 AM, [e-mail] wrote:
> >
> > Hi all,
> >
> > I would like to share with this group a short document that explains how
> PCP can be of great use in the context SIP-based services:
> > http://tools.ietf.org/html/draft-boucadair-pcp-sip-ipv6-03
> >
> > As indicated in the I-D, the main benefits include (but not limited to):
> >
> >    o  Avoid embedding an ALG in the middleboxes.  Note, ALGs are not
> >       recommended since the evolution of the service would depend on the
> >       ALG maintenance.
> >    o  Not require any Hosted NAT Traversal function (e.g., [RFC7362]) to
> >       be embedded in the SIP server.  Intermediate NATs and firewalls
> >       are transparent to the SIP service platform.
> >    o  Avoid overloading the network with keepalive message to maintain
> >       the mapping in intermediate middleboxes.
> >    o  Work without requiring symmetric RTP/RTCP [RFC4961].
> >    o  Not require symmetric SIP to work (i.e., rport [RFC3581]).
> >    o  Easily support unidirectional sessions.
> >
> > When this document was first presented in the PCP WG, I was suggested
> that it is better to publish it in RAI with a review from the PCP WG.
> Hence, this message to the list.
> >
> > Cheers,
> > Med
> > _______________________________________________
> > dispatch mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/dispatch