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

Chapter 191
Dynamic Port Numbers

TCP/IP also implements dynamic port numbers. Since the Port Number field in the UDP header is 16 bits long, 65,535 ports (minus the assigned port assignments) are available for individual use. This range can be used for registered and dynamic ports. One use for a dynamic port is as a source station that is requesting the services of TFTP on a remote station. The source station dynamically assigns itself an available port number (usually above 1024 to use so that the remote station knows what port to access when it transfers the file). In other words, if a user initiates a trivial file transfer (TFTP), the TFTP request packet sent to the TFTP server includes in its UDP header a dynamic port number of the requesting network station that wanted the TFTP, called the source port. Let’s say it is assigned port 2000. The destination port number would be 69. In this way, the server will accept the packet, give it to the TFTP process in the host and, when the host responds, it will know how to address the port number in the response packet. In the response packet, the server would fill out the UDP header with a destination port of 2000, source port of 69, and send the packet back to the requesting station.

Another use is when network vendors implement proprietary schemes on their devices; for example, a proprietary scheme for a network station to boot or a proprietary scheme to allow network management statistics to be gathered. All these applications are valid and may run on any TCP environment using a dynamic port assignment.

Dynamic Port Numbers

The disadvantage of dynamic ports occurs when a broadcast IP datagram is transmitted to the network using a dynamic port. This port could be used by another vendor on the network, and another network station may invoke a process to accommodate that request. This is rare, but has been known to happen.

Dynamic port numbers are assigned by the TCP/IP software at the local workstation, and can be duplicated from workstation to workstation without respect to the application. This is because an application on any network station is uniquely identified by the IP address (network number, host number) and the port number. When taken as a whole number, it is called the socket number, and cannot be duplicated on an IP network except by negligence.

Final note: Some people like to use the terms port and socket interchangeably. You can, but proper IP semantics state that a port number and a socket number are not the same thing, as indicated in the preceding paragraph.