On 13 April 2011 09:44, Martin Sustrik <[email protected]> wrote: > On 04/13/2011 09:26 AM, Martin Sustrik wrote: > > > Ah, I should have said IP+DNS rather than just IP. Sorry. > > I would suggest looking at the design of HTTP, which is pretty similar > to the "send-to-address" pattern we are discussing. > > HTTP is successful because it doesn't try to reinvent the wheel. It uses > DNS to resolve the name of the site rather than managing a global > namespace of its own. Then it uses IP (via TCP) to route the message to > the destination. It's not concerned with nodes in the middle. It treats > the connection to the host as a single hop, leaving the details of > routing over the Internet to L3.
HTTP is a good example of multiplexing based on DNS name rather than IP, using the "host:" request header, which interestingly, if I remember well, was regulated as mandatory in the 1.1 version of the protocol. If zeromq had some handshake where the connecting socket could say "I actually want to connect somedomain:5123", the same receiving tcp socket could be used to spawn different connections to different zeromq sockets. The application could simply bind several zeromq sockets on the same ip/port, providing that they have different 'virtual hosts', and zmq will attach new connections to the correct sockets after reading the 'vhost' label in the first frame. Sorry, perhaps I entered late in the whole thing, but I have the impression that there is no such thing right now. I guess it could be transparently implemented with a transport (vtcp?) Cheers, Marko
_______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
