English
UDP uses a single segment format shown in the figure below.
UDP Header Format
This limitation is due to the fact that the network layer cannot transport packets that are larger than 64 KBytes. As UDP does not include any segmentation/reassembly mechanism, it cannot split a SDU before sending it. The UDP header consumes 8 bytes and the IPv6 header 60. With IPv4, the IPv4 header only consumes 20 bytes and thus the maximum UDP payload size is 65507 bytes.
The User Datagram Protocol (UDP) is defined in :rfc:`768`. It provides an unreliable connectionless transport service on top of the unreliable network layer connectionless service. The main characteristics of the UDP service are :
The User Datagram Protocol
the UDP service will not deliver a corrupted SDU to the destination
the UDP service does not guarantee the delivery of SDUs (losses can occur and SDUs can arrive out-of-sequence)
the UDP service cannot deliver SDUs that are larger than 65467 bytes [#fmtuudp]_
The UDP header contains four fields :
the registered port numbers (officially 1024 <= port < 49152)
the privileged port numbers (1 < port < 1024 )
The figure below shows a typical usage of the UDP port numbers. The client process uses port number `1234` while the server process uses port number `5678`. When the client sends a request, it is identified as originating from port number `1234` on the client host and destined to port number `5678` on the server host. When the server process replies to this request, the server's UDP implementation will send the reply as originating from port `5678` on the server host and destined to port `1234` on the client host.
the ephemeral port numbers ( officially [#fephemeral]_ 49152 <= port <= 65535 )
the entire UDP segment, including its header
The complete list of allocated port numbers is maintained by IANA_ . It may be downloaded from http://www.iana.org/assignments/port-numbers
The checksum of the UDP segment is computed over :
Several types of applications rely on UDP. As a rule of thumb, UDP is used for applications where delay must be minimized or losses can be recovered by the application itself. A first class of the UDP-based applications are applications where the client sends a short request and expects a quick and short answer. The :term:`DNS` is an example of a UDP application that is often used in the wide area. However, in local area networks, many distributed systems rely on Remote Procedure Call (:term:`RPC`) that is often used on top of UDP. In Unix environments, the Network File System (:term:`NFS`) is built on top of RPC and runs frequently on top of UDP. A second class of UDP-based applications are the interactive computer games that need to frequently exchange small messages, such as the player's location or their recent actions. Many of these games use UDP to minimize the delay and can recover from losses. A third class of applications are multimedia applications such as interactive Voice over IP or interactive Video over IP. These interactive applications expect a delay shorter than about 200 milliseconds between the sender and the receiver and can recover from losses directly inside the application.
In most Unix variants, only processes having system administrator privileges can be bound to port numbers smaller than `1024`. Well-known servers such as :term:`DNS`, :term:`NTP` or :term:`RPC` use privileged port numbers. When a client needs to use UDP, it usually does not require a specific port number. In this case, the UDP implementation will allocate the first available port number in the ephemeral range. The range of registered port numbers should be used by servers. In theory, developers of network servers should register their port number officially through IANA [#fportnum]_, but few developers do this.
Footnotes
Computation of the UDP checksum