Ah, yes, brokered UDP traversal of a NAT. Bryan Ford of MIT describes this
here: http://www.pdos.lcs.mit.edu/~baford/nat/draft-ford-natp2p-00.txt and
here: http://www.pdos.lcs.mit.edu/~baford/nat/. The technique requires the
NAT devices keep a constant mapping between public and private address/port
pairs, and do *not* allocate new port mappings for different
destination/port pairs. (This is sometimes known as "full cone NAT".)

SIP has various modifications to allow a session description to distinguish
between public and private addresses, so that it will effectively broker
the direct connection between the two endpoints. Presumably Packet8's
service does this. As for Skype, presumably their proprietary call
signaling protocol does something equivalent.

The alternative is to use something called a Session Border Controller,
which sits in the public address space, does address mapping and acts a
relay for the media stream. However, these devices are not cheap
(requiring as they do a great deal of packet forwarding capacity for VoIP),
and the media streams have to go via the VoIP service provider's bandwidth,
which adds to the cost.

On the other hand, if the FBI get their way under the CALEA regulations, a
VoIP service provider will have to be able to tap the voice stream, and in
a manner undetectable to the wiretap subjects. That might mandate using
SBCs, as I believe Vonage (for example) does.

The fact that Skype is P2P and encrypted is probably giving the Feds heartburn.

Rob

(robert at welbourn dot com)

At 10:07 AM 9/3/2003 -0500, brandon@... wrote:
> > Regarding Packet8's 'magic'... could it be something like the next-port
> > tricks that are possible with some NATs, i.e. A and B use a relay to
> > co-ordinate their timing and port numbers so they each initiate
> > connections to the other's NAT at the same time on port+1 and rely on
> > the fact that the other's NAT will (probably) assign the next sequential
> > port to the outgoing connection. If it works, A and B can then talk
> > directly even though they're mutually NAT'd. What else could it be?
>
>Well my box arrives on Friday, so I'll let you know.
>
>
>To unsubscribe from this group, send an email to:
>decentralization-unsubscribe@egroups.com
>
>Announce or discover P2P conferences on the P2P Conference Wiki at
>http://www.neurogrid.net/twiki/bin/view/Main/PeerToPeerConferences
>
>Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/