Conceptually, the TCP/IP protocol stack consists of four layers, each layer consisting of one or more protocols. A protocol is a set of rules or standards that two entities must follow so as to allow each other to receive and interpret messages sent to them. The entities could, for example, be two application programs in an application protocol, or the entities might be two TCP protocol layers in two different IP hosts (the TCP protocol).

Figure 1 illustrates the TCP/IP protocol stack.

Figure 1. The TCP/IP protocol stack

Programs are located at the process layer; here they can interface with the two transport layer protocols (TCP and UDP), or directly with the network layer protocols (ICMP and IP).

TCP

Transmission Control Protocol is a transport protocol providing a reliable, full-duplex byte stream. Most TCP/IP applications use the TCP transport protocol.

UDP

User Datagram Protocol is a connectionless protocol providing datagram services. UDP is less reliable because there is no guarantee that a UDP datagram ever reaches its intended destination, or that it reaches its destination only once and in the same condition as it was passed to the sending UDP layer by a UDP application.

ICMP

Internet Control Message Protocol is used to handle error and control information at the IP layer. The ICMP is most often used by network control applications that are part of the TCP/IP software product itself, but ICMP can be used by authorized user processes as well. PING and TRACEROUTE are examples of network control applications that use the ICMP protocol.

IP

Internet Protocol provides the packet delivery services for TCP, UDP, and ICMP. The IP layer protocol is unreliable (called a best-effort protocol). There is no guarantee that IP packets arrive, or that they arrive only once and are error-free. Such reliability is built into the TCP protocol, but not into the UDP protocol. If you need reliable transport between two UDP applications, you must ensure that reliability is built into the UDP applications.

ARP/ND

The IPv4 networking layer uses the Address Resolution Protocol (ARP) to map an IP address into a hardware address. In the IPv6 networking layer, this mapping is performed by the Neighbor Discovery (ND function). On local area networks (LANs), such an address would be called a media access control (MAC) address.

RARP

Reverse Address Resolution Protocol is used to reverse the operation of the ARP protocol. It maps a hardware address into an IPv4 address. Note that both ARP packets and RARP packets are not forwarded in IP packets, but are themselves media level packets. ARP and RARP are not used on all network types, as some networks do not need these protocols.