| The conversion of addresses between different protocol layers represents an important task for unique identification of resources in a computer network. Such a conversion is required at the transition between two neighbouring layers within a reference model, because each layer uses its own address types, depending on its functionality (IP, MAC, ATM addresses, etc.). For example, the destination computer is specified in the form of an IP address if a packet is sent over the Internet Protocol. This address is valid only within the IP layer. In the data-link layer, both the service used by the Internet Protocol to transport its data and different LAN technologies (e.g., Ethernet, token ring, ATM), each with its own address formats, can be used. The network adapters of a LAN are generally identified by 48-bit addresses, so-called MAC addresses. A MAC address identifies a unique network adapter within a local area network. To be able to send a packet to the IP instance in the destination computer or to the next router, the MAC address of the destination station has to be determined in the sending protocol instance. The problem is now to do a unique resolution of the mapping between a MAC address and an IP address. What we need is a mapping of network-layer addresses to MAC addresses, because the sending IP instance has to pass the MAC address of the next station in the form of interface control information (ICI) to the lower MAC instance. (See Section 3.2.1.) At the advent of the Internet, this mapping was implemented by static tables that maintained the mapping of IP addresses to MAC addresses in each computer. However, this method turned out to be inflexible as the ARPANET grew, and it meant an extremely high cost when changes were necessary. For this reason, RFC 826 introduced the Address Resolution Protocol (ARP) to convert address formats. Though the TCP/IP protocol suite has become the leading standard for almost all computer networks, it is interesting to note that ARP was not designed specifically for mapping between IP and MAC addresses. ARP is a generic protocol that finds a mapping between ordered pairs (P,A) and arbitrary physical addresses, where P is a network-layer protocol and A is an address of this protocol P. At the time at which ARP was developed, different protocols, such as CHAOS and Decnet, had been used in the network layer. The ARP instance of a system can be extended so that the required addresses can be resolved for each of the above combinations, which means that no new protocol is necessary. The most common method to allocate addresses between different layers maps the tuple (Internet Protocol, IP address) to 48-bit MAC addresses. |
