II.2.3-Architectural model
We saw that the goal of an internet is to provide universal interconnection (while allowing individuals to use whatever network hardware is best suited to their needs).
First, how is interconnection possible between two networks using heterogeneous architectures?
Two such networks are connected through a special (and usually dedicated) computer called gateway or router.
This computer is physically connected to both networks. It reads data coming through one of its two network interfaces, and (if the destination of this data is on the other network) it puts it on the other network through its second network interface.
fig. 2.9: example of interconnection between an Ethernet and a Token ring network
But such interconnection is not enough to make an internetwork. We also want the topology of the internet and its constituting elements to be free. For example, a new network being connected to an internet should not need to be connected to every other network. This implies that data should be able to cross one (or more) intermediate networks before reaching its ultimate destination.
As a consequence, to be able to route data across the internet, gateways needs to know the topology of the internet. Routing is a complex problem, and an internet makes it even more complex. We will see how it was somewhat simplified by routing data according to network instead of host destination, thus making routing tables much less bulky.
To provide universal service, the Internet uses a scheme that hides the details of underlying network hardware.
Interconnection could be provided at application level, but we saw in the previous section how difficult it is to maintain such a system for a simple network. A complex internetwork, where application programs will have to run on widely different hardware architectures, makes this solution even less attractive.
Instead, interconnection is provided at network level. Data communication activities are then separated from applications, making the system more flexible and easier to adapt to changes, either in network architecture or in application programs. Of course, the work of the application programmer is also made much easier.
Network level-interconnection provides universal interconnection, and makes our heterogeneous collection of networks appear as a single large packet switching network to the user, where the gateways would be the switches (cf. figure 2.10). This is what is properly called an internet.
The use of protocol layering and more precisely of the protocol suite called TCP/IP made this possible.