The Basics of IP Transport

As shown in Figure 8-1, IP corresponds to layer 3 of the Open Systems Interconnection
(OSI) seven-layer protocol stack. At its most basic level, IP simply
passes a packet of data from one router to another through the network
to the appropriate destination, as identified by the destination IP address
in the IP packet header. This simple operation means that IP is inherently
unreliable. IP provides no protection against a loss of packets, which might
happen if congestion occurs along the path from the source to the destination.
Moreover, in a given stream of packets from the source to the destination,
it is quite possible that packets will take different routes through the network, meaning that different packets can have different delays and also
that packets may arrive at the destination out of sequence.
In data networks, in order to ensure an error-free, in-sequence delivery
of packets to the destination application, the Transmission Control Protocol
(TCP) is used. This protocol resides on the layer above IP. When a session is
to be set up between two applications, the application data is first passed to
TCP where a TCP header is applied, the data is then passed to IP where an
IP header is applied, and then it is forwarded through the network. The
information contained in the TCP header includes, among other things,
source and destination port numbers, which identify the applications at
each end; sequence numbers and acknowledgement numbers, which enable
the detection of lost packets; and a checksum, which enables the detection
of corrupted packets. TCP uses these information elements to request
retransmission of lost or corrupted packets and to deliver packets to the
destination application in the correct order. In order to do all of this, TCP
first establishes a connection between peer TCP instances at each end. This
involves a sequence of messages between the TCP instances prior to the
transfer of user data.
Instead of using TCP at layer 4 in the stack, the User Datagram Protocol
(UDP) is another option. This is a simple protocol, which does little more
than enable the identification of the source and destination applications. It
does not support recovery from loss or error and does not ensure an insequence
delivery of packets. It is meant for simple request-response types
of transactions, rather than the sequential transfer of multiple packets.An
application that would use UDP rather than TCP, for example, is the
Domain Name Service (DNS), a classic one-shot request-response protocol.