> (2) There is no way that these decisions can be made solely at the > transport or application layer, because source (and to a lesser degree > destination) address selection is tightly tied to the first-hop > forwarding decision. The outbound interface, source address and default > router all have to be selected in a coordinate process, to avoid sending > traffic that will be discarded on the outbound path, due to router > filters. Actually, applications can to do that today, using the socket API, if the stack implements the "strong host" model. The application just needs to bind the socket to a specific IP address. Doing that ensures that packets sourced by the application will use the specified address, will go out through the interface corresponding to that address, and will use the default gateway associated with that interface. -- Christian Huitema _______________________________________________ Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf