Illustrated TCP/IP by Matthew G. Naugle Wiley Computer Publishing, John Wiley & Sons, Inc. ISBN: 0471196568   Pub Date: 11/01/98

Chapter 189
Port Numbers

Review RFC 814. Since many network applications may be running on the same machine, a method is needed to allow access to these applications, even though they reside on the same machine and the machine contains one IP address. One IP address and many applications? How do we decide which datagram belongs to which application?

It would not be advantageous to assign each process an IP address, nor to change the IP addressing scheme to include a marker to identify a unique application in the machine. Instead, both the TCP and UDP protocols provide a concept known as ports (sometimes mistakenly called sockets, which is not correct). Ports, along with an IP address, allow any application in any machine on an internet to be uniquely identified.

There are three different types of port numbers: assigned, registered, and dynamic. The RFC of assigned numbers (RFC 1700 at the time of this writing) contains assigned and registered numbers. The first 1024 ports are assigned and in specific use and should not be used by any application. The remaining addresses can be dynamic and registered (16 bits allows for 65,535 ports) and can be used freely, although IANA does request that vendors register their application port numbers with them.

When a station wishes to communicate to a remote application, it must identify that application in the datagram. For example, if a station needed to use a simple file transfer protocol known as trivial file transfer program (TFTP) on the station 130.1.1.1, it would address the datagram to station 130.1.1.1 and insert destination port number 69 in the UDP header. The source port number identifies the application on the local station that requested the file transfer, and all response packets generated by the destination station would be addressed to that port number on the source station. Generally, the source port is randomly generated by the source station. If the source port is not used (broadcast RIP update tables), it should be set to 0. So, when the IP layer demultiplexes the packet and hands it to UDP, UDP will pass the data to the locally assigned port number for it to process the data.