Things about UDP protocol

UDP (User Datagram Protocol) protocol, translated as User Datagram Protocol, is located in the transport layer of the OSI model, just like TCP protocol. However, compared with TCP protocol, UDP protocol is much simpler, because it does not have complex processing mechanisms such as "flow control" and "congestion control". It does not even have a retransmission mechanism, that is to say, if your data packet is lost halfway, it is really impossible to find it back, so UDP protocol is unreliable. Of course, this retransmission mechanism is for the transport layer, you can write a protocol at the application layer to handle packet loss, such as adding ACK and sequence number mechanisms like TCP.

Then you may wonder, why choose UDP protocol instead of reliable TCP protocol?

UDP segment structure

This of course depends on the needs of the application, but before talking about this topic, let's first take a detailed look at the UDP protocol.

To be honest, the UDP segment structure is much simpler than the TCP segment (see the figure below). After all, the UDP protocol does not have any redundant mechanisms.

Let's get back to the point. The "source port number" and "destination port number" in the message segment are used to tell the transport layer where (which process) the message comes from and where (which process) it is going to. But please note that a UDP socket is identified by a two-tuple, which refers to the destination IP address and the destination port number. In other words, the corresponding process on the server does not care which client you come from. I will put it into the same socket for processing. After processing, I will send the response information to the client based on the source port number and source IP address. In comparison, a TCP socket requires a four-tuple to identify: source IP address, source port number, destination IP address and destination port number. This will be discussed in detail when talking about the TCP protocol, so I will not repeat it here.

PS: You may ask, why is there no IP address in this message segment? This is because the IP address is stored in the IP protocol segment of the network layer, and of course it is not in the message segment of the transport layer.

No connection

Whenever we mention the TCP protocol, the first thing that comes to our mind is the three-way handshake and four-way wave. For the UDP protocol, these are not the case. When using the UDP protocol, if the client wants to send a message segment to the server, it does not need a handshake and can be sent directly. For this reason, the UDP protocol is called connectionless.

It is easy to imagine that if the handshake process is not required, there will be no delay caused by establishing a connection. In a word, fast! This is also a big part of the reason why DNS (Domain Name System) runs on the UDP protocol.

But the UDP protocol is unreliable. What should we do if packets are lost during transmission? The simplest way is to ignore it! (Otherwise, as mentioned at the beginning of the article, a retransmission mechanism must be implemented at the application layer.) Take DNS for example. If a data packet is lost, the client will resend it (with a timeout mechanism), and under normal circumstances, the probability of packet loss is very low. However, if the TCP protocol is used, domain name query will be much slower because a connection must be established. In addition, the network overhead of the UDP protocol is smaller - the UDP segment has an 8-byte header overhead, while the TCP protocol has a 20-byte overhead (see the two previous pictures about the segment). Small network overhead means that the DNS server can accept more client requests.

Another aspect is that the TCP protocol has a congestion control mechanism, which will curb the TCP sender when the network is congested, thereby delaying the transmission of the message segment. Therefore, for some real-time programs that require low transmission delay and can tolerate some data loss, the UDP protocol may be a better choice. The Routing Protocol (RIP) and the Network Management Protocol (SNMP) also choose UDP as the underlying transmission protocol.

Finally, here is a flow chart of the client and server communicating using the UDP protocol:

There is not much to say about the UDP protocol. The TCP protocol to be discussed next time is more brain-burning, so be prepared!