zweig@m.cs.uiuc.edu (05/31/89)
Since a gateway may have two IP address -- say 100.100.100.100 for one network and 101.101.101.101 for the other network it's connected to -- I don't see what the "right way" to describe an endpoint of a TCP connection is. To read RFC 793 very literally, it would seem that port <100.100.100.100,42> ought to specify a different socket/enpoint than <101.101.101.101,42>. But they obviously both refer to "TCP port 42" on the gateway machine.... If the answer is "different", would any implementations that assumes they're the same because there is only one machine involved please raise their hands? If the answer is "same", would any implementations that assumed than different <IP-address, port-#> pairs need to specify different endpoint please say so? -Johnny Confused
ron@ron.rutgers.edu (Ron Natalie) (06/01/89)
> To read RFC 793 very literally, it would seem that port <100.100.100.100,42> > ought to specify a different socket/enpoint than <101.101.101.101,42>. But > they obviously both refer to "TCP port 42" on the gateway machine.... They are differnt. TCP connections are identified by the host address and port on both ends of the connection. Saying TCP port 42 is meaningless, it is only valid in the context of the local IP address and the remote address/port pair. For example 100.100.100.100,42 101.101.101.101,42 100.100.100.100,42 200.200.200.200,42 200.200.200.200,42 101.101.101.101,42 refer to three seperate TCP connections without ambiguity.
CERF@A.ISI.EDU (06/09/89)
The two sockets (on different IP addresses) should be considered distinct. Vint Cerf