Q: Why does TCP/IP need its own addresses?

A: The TCP/IP protocols were designed from the start to support all network-capable computing platforms. On any computer network, there must be some means of uniquely identifying specific machines. The Ethernet and Token Ring systems that are now used on the majority of local area networks have their own unique addresses encoded at the factory, but this is not true for all computers. The TCP/IP standards include their own addressing system to ensure that any computer connected to the Internet (or to any other internetwork) can be distinguished from all of the others.

TCP/IP addressing is the responsibility of IP, and occurs at the network layer of the protocol stack. The network layer protocol is also responsible for routing packets to destinations on other networks. As a result, the destination IP address in the header of every IP datagram identifies the system that is to be the ultimate recipient of the packet, even if it is located on another network halfway around the world.

Data link layer protocols like Ethernet, however, are not conscious of systems on other network segments. The destination address in an Ethernet packet header identifies the next system on the local network segment that is to receive the packet, whether that system is the actual destination specified in the IP header or a router that provides access to other networks.

Therefore, while it may seem as though having two addresses on one system is redundant, the network layer and data link layer addresses actually perform different functions. Both are needed for a TCP/IP internetwork to function properly.

For more information, see page 289 of TCP/IP Administration.