[LINK] www.ipv6.org.au/summit

[stephen at melbpc.org.au]

Karl writes,

> You're missing the point. The original message proposed the rather sad,
> small-visioned and faintly absurd idea of stealing some of the port bits
> in the IP packet to extend the address size. That was "supported" by the
> statement that it was not anticipated that broadband consumers would
> need many addresses. The above is my response. If you build a world with
> low ceilings don't expect to see anyone invent flying machines. Regards K

The research ideas you call 'sad & absurd' seem to be flying well, Karl :-)

 "A Better Approach than Carrier-Grade-NAT"

A paper by: Olaf Maennel, Randy Bush, Luca Cittadini and Steven Bellovin, 
from T-Labs, IIJ, Universita’ Roma-Tre and Columbia University respectively
 
 http://rip.psg.com/~randy/080820.alt-to-cgn.pdf

(continuing quote from previous email ..)

. There is a long history of treating port numbers as part of the network
address. It was considered as part of the design of TCP/IP. In the same
time frame, Pup and the Xerox Network System architecture included
the “socket number” as part of an address; the other two parts were a
network number and a -bit host number. However, only the network number
was used for routing. Later, Bellovin and colleagues made suggestions that
embedded the service in the IP address; In it, Hang Zhao, Chi-Kin Chau,
and Steven M. Bellovin suggest routing on the !address,port"/48 string,
i.e., a route per service. Thus, if there is no route advertisement for,
say, !A,25"/48, every router along the path will decline to forward SMTP
packets to host A. However, that work had no notion of address or port
number translation. This work was partially supported by a gift from Cisco.
--


Cheers people
Stephen Loosley
Victoria Australia