Michel; > In terms of design, if you do TCP/IP *only* design, the TCP/IP model is > probably enough. However, the Internet is not only TCP/IP. Carriers, for > example, don't care much if their fiber transports TCP/IP or IPX or > voice or video or GigE. No. Anything at or above transport layer is a layer internal to end systems and has nothing to do with networking or network protocols. Seperation of transport and application layers is a overkill for a best effort network, though it may help standardize the internal design of end systems such that anything supported by kernel belong the transport layer. You can check the reality that application and transport areas of IETF are now almost identical, though, historically, trasnsport area was working on protocols likely to be implemented in kernel. In addition, defining a thin transport layer may be useful over a hypothetical port-number-aware network such as that supporting RSVP. However, forcibly defining a session-layer-aware network is a layer violation. > And, there are complex multi-protocol networks that a) don't use only > TCP/IP and b) would not be able to use the TCP/IP model anyway because > it's too simple. Unless you are trying to standardize internal design of application layer gateways, which is like defining standardizing the way of structured programming and is hopeless, the separation of upper layers is meaningles. > The bottom line is: lots of people are going to continue using the OSI > model. We don't need two different models. I am having no difficulty in teaching my students, even though I often forget the names of two OSI layers between transport and application. In writing this mail, I only remember one: session. New comers don't need two different models. Masataka Ohta