TCP and UDP

The Transmission Control Protocol (TCP) and User Datagram Protocol (UDP) are the two most popular protocols in the transport layer. They ensures that messages are delivered error-free, in sequence, and with no losses or duplication. The key difference between TCP and UDP is that TCP provides a wide variety of services to applications, whereas UDP does not. At the result of this, TCP is much more complex than UDP so this tutorial is dedicated to explore TCP in detail but we still compare them.

Both TCP and UDP are protocols at the Transport layer (of both OSI and TCP/IP model) but why we need both of them? The answer is:

+ TCP is slower but reliable
+ UDP is faster but unreliable

In most cases we will want to be reliable in web accessing, email communicating, file uploading… as we don’t expect a few corrupted packets would destroy our whole work. With TCP, these corrupted packets will be resent or repaired to make sure everything is correct. Yes, TCP is really nice to ensure your work is accurate!

But with a price…

To guarantee the sending segments is free of error, TCP adds some bits for tracking and checking purpose so that the other end can verify and ask for missing pieces of segments. As a result of this, the segments become larger, consume more bandwidth and CPU resources to proceed.

Although UDP cannot guarantee everything is accurate like TCP but UDP is faster than TCP because it does not require additional bits for tracking and checking purpose. So which tasks need speed? Video (streaming) and audio are ideal for this task because they are considered real-time applications. Suppose you are talking to your friend, surely you want your voice to reach your friend without any delay. It would be very weird if your friend can only hear your voice after a few seconds.

Note: Segment is the name of the data packet at Transport layer

TCP can also slow down the transmission if it sees the path to the destination is too crowded. You don’t want TCP to slow down your voice in traffic-jam hours either. For example when you say “Hello, how are you?”, your friend at the other end may hear “Hellooooo,…… hooooooooow arrrrrrrre yyyyyoou”. What is an awful conversation!

Losing a few packets for voice or video is acceptable. For example if you say the word “Hello” in one second, an IP phone generates about 25 to 100 packets (just an estimation, it depends on the codec and sampling frequency) so your friend can still understand what you say even if a few packets are missing. Moreover, re-transmission the missing packets is not useful as voice and video are real-time applications and the receiving end cannot wait for the missing segments to be resent.

So now we have some basic understanding of TCP and UDP. In the next part we will learn more about TCP. Let’s start with how TCP set up and terminate a connection.

TCP three-way handshake (to start the communication)

Suppose host A wants to start communicating with host B using TCP. Before they can send real data, a three-way handshake must be established first. Let’s see how this process takes place:

1. First host A will send a SYN message (a TCP segment with SYN flag set to 1, SYN is short for SYNchronize) to indicate it wants to setup a connection with host B. This message includes a sequence (SEQ) number for tracking purpose. This sequence number can be any 32-bit number (range from 0 to 2³²) so we use “x” to represent it.

2. After receiving SYN message from host A, host B replies with SYN-ACK message (some books may call it “SYN/ACK” or “SYN, ACK” message. ACK is short for ACKnowledge). This message includes a SYN sequence number and an ACK number:
+ SYN sequence number (let’s called it “y”) is a random number and does not have any relationship with Host A’s SYN SEQ number.
+ ACK number is the next number of Host A’s SYN sequence number it received, so we represent it with “x+1”. It means “I received your part. Now send me the next part (x + 1)”.

The SYN-ACK message indicates host B accepts to talk to host A (via ACK part). And ask if host A still wants to talk to it as well (via SYN part).

3. After Host A received the SYN-ACK message from host B, it sends an ACK message with ACK number “y+1” to host B. This confirms host A still wants to talk to host B.

If you are still unclear about this process, let’s assign: x = 1 and y = 50:

In this process, three messages need to be sent so we often call it “three-way handshake”.

Nice, now you really understand TCP three-way handshake, right? Host A can start sending real traffic to host B after the three-way handshake process.

TCP also does nearly the same thing when one end wants to terminate the connection with TCP four-way termination process.

TCP four-way termination (to end the communication)

Suppose Host A wants to end the connection to host B, Host A will send a FIN message (a TCP segment with FIN flag set to 1), FIN is short for FINISH. The purpose of FIN message is to enable TCP to gracefully terminate an established connection. Host A then enters a state called the FIN-WAIT state. In FIN-WAIT state, Host A continues to receive TCP segments from Host B and proceed the segments already in the queue, but Host A will not send any additional data.

Device B will confirm it has received the FIN message with an ACK (with sequence x+1). From this point, Host B will no longer accept data from Host A. Host B can continue sending data to Host A. If Host B does not have any more data to send, it will also terminate the connection by sending a FIN message. Host A will then ACK that segment and terminate the connection.

TCP requires to establish and terminate the connection before and after exchanging real traffic so it is called connection-oriented protocol. UDP does not support these features so it is called connectionless protocol.

More formally, these terms can be defined as follows:
+ Connection-oriented protocol: requires a logical connection to be established between the two processes before data is exchanged
+ Connectionless protocol: allow data to be exchanged without setting up a link between processes

In conclusion, TCP requires the establishment (via three-way handshake) and termination (via four-way termination) of a connection. In the next part we will learn about popular TCP features.

TCP Features

Some popular TCP features we will learn here are: Multiplexing using port numbers, Flow control using windowing and Reliability (Error Detection and Error recovery)

Multiplexing using port numbers

Suppose you are using a laptop for web browsing, email communicating and FTP uploading at the same time. All of them require using TCP while your laptop only has one IP address (with one network card) so how your laptop knows which packets received from the Internet are dedicated for which application?

Above question is solved with port numbers. Each application will use a different and available port number to communicate with outside world. For example your laptop can choose port 50000 for web browsing, port 50001 for email communicating and port 50002 for FTP uploading.

Notice that your laptop can choose any available source port but it must use pre-defined destination ports for well-known services. Port numbers are defined in three ranges:
+ Well-known port numbers (0 through 1023): assigned to key or core services that systems offer
+ Registered port numbers (1024 through 49151): assigned to industry applications and processes. For example: 1433 is assigned for Microsoft SQL Server process)
+ Dynamic port numbers (49152 through 65535): used as temporary ports for specific communications. Our laptop can use these ports for communication

The table below lists TCP ports for well-known services:

TCP Service	Description	Port
FTP	File Transfer Protocol	20, 21
SSH	Secure shell	22
Telnet	Terminal network	23
SMTP	Simple Mail Transfer Protocol	25
DNS	Domain Name Server	53
HTTP	Hyper Text Transfer Protocol	80
HTTPS	Hyper Text Transfer Protocol Secure	443

Note: There are some other well-known ports that are not listed here. The well-known ports are assigned by the Internet Assigned Numbers Authority (IANA) in the range of 0 to 1023.

Multiplexing relies on a concept called a socket. A socket consists of three things:

+ An IP address
+ A transport protocol
+ A port number

So suppose the IP address on our laptop is 123.1.1.1 and use TCP to access web server with port 50000, we may write the socket (123.1.1.1, TCP, 50000). For web server application running on the Web Server with IP 200.1.1.1 the socket should be (200.1.1.1, TCP, 80) as the web server uses the well-known port 80 for HTTP.

The socket on each computer is unique so the connection between two sockets on two computers identify a unique connection between them. Therefore you can use multiple applications on the same computer at the same time. Each application will use a unique source port so they cannot interfere with each other.

We only mentioned about source ports but notice TCP header requires both source port and destination port. That means if our laptop wants to connect to a Web Server it must include the destination port in TCP header as well. The destination port for Web Server in this case is 80. When the Web Server replies to our laptop, it uses the laptop’s source port as its destination port (50000 in this case).

Note: Both TCP and UDP use multiplexing with port numbers for their services.

Flow-control using windowing

In the TCP header there is a field called “Window” which plays an important role in the TCP transmission. A “Window” specifies the number of segments the sender can forward without receiving an acknowledgment. It is the key to transfer data and flow control efficiently. Let’s see how it works!

After the TCP connection has been established, both the client and server use this Window field to tell the other how many bytes of data it is willing to receive at one time before sending an acknowledgement to the sender. The larger the window size number (in bytes), the greater the amount of data that the host can transmit. For example, with a Window size of 1 (byte), every one byte must be acknowledged before sending the next one.

But waiting for ACK after each segment would be very inefficient. So TCP tries to increase the Window size to 3 (bytes), which means every three bytes can be received before sending the acknowledgement.

As you can see, the bigger the Window size, the fewer ACKs needed to be sent and the more efficient the transmission is. So the receiver will try to increase the Window size after each successful transmission so that the sender can send more. But the Window size cannot increase forever, TCP stops increasing the Window size when the receiver does not send an ACK (within a specific time period) or when the Window size reaches its maximum value. If a congestion occurs on the link then TCP may decrease the Window size.

The window size is variable during the lifetime of a connection so we often refer it as a “sliding window”.

If the sender does not receive the ACK in time, it knows that the segments should be resent, and that the transmission rate should be slowed down. Suppose Host A did not receive the expecting ACK 7 then it knows segments 4, 5, 6 should be resent.

Reliability (Error Detection and Error recovery)

This is the most important feature of TCP. TCP must recover from data that is damaged, lost, duplicated during the transmission. But please grasp the difference between error detection and error recovery first:
+ Error detection: the detection of errors during the transmission. Error detection does not repair corrupted data, it just detects it
+ Error recovery: the detection of errors and repair them

To achieve error detection, TCP adds some extra bits to the data, called checksum. A TCP sender computes the checksum value based on the contents of the TCP header and data fields. This 16-bit value will be compared with the value the receiver generates using the same computation. If the values match, the receiver can believe that segment arrived intact. If the values do not match, the receiver indicates an error occurred and the segment is discarded and a notification will be sent to the receiver depending on how the TCP stack is implemented on the receiver’s operating system.

To achieve error recovery, TCP uses the Sequence number (at the sender’s side) and Acknowledgement fields (at the receiver’s side) in the TCP header. These two fields are also used to find out lost, duplicated segments. Let’s see an example.

In the transmission below, host A sends three segments 1, 2, 3 to host B. Segment 2 was lost while segment 3 arrived to Host B. Then Host B replied with an ACK 2, implying that it is expecting segment 2 next. Host A can re-send another segment 2 to recover the lost segment. If Host B receive that segment it will ask for the segment 4 (because it already has segment 3).

Error recovery

You may ask “what will happen if the ACK 2 sent from Host B is also lost?” In fact, after sending each segment Host A sets a retransmission timer, just in case the ACK is lost (or all the sending segments are lost; Host B would not send ACK in this case because it did not receive anything). If this timer expires, Host A will send all the segments again.

Note: UDP does support error detection (via checksum) but it does not support error recovery. If UDP finds a corrupted segment, it just simply drop it.

Let’s sum up all things we learned about TCP and UDP so far.

Same:

+ Both TCP and UDP operate at Transport Layer
+ Both TCP and UDP use Multiplexing via port numbers

Difference:

TCP	UDP
Reliable	Unreliable
Connection-oriented	Connectionless
Segment retransmission and flow control through windowing	No retransmission or windowing
Segment sequence	No sequencing
Acknowledge segment	No acknowledgement
Start and end the communication by three-way handshake and four-way termination	No action is required before and after sending real data

Finally we show the TCP and UDP header in detail for your reference. There are some fields which are out of scope of this tutorial.

TCP Header (20 bytes)

Notice about the FLAG fields (between the “Reserved” and “Window Size” fields). If SYN bit is turned on, it is a SYN message. If ACK bit is turned on, it is an ACK message. If both SYN and ACK bits are turned on, it is a SYN-ACK message.

And this is the UDP header:

UDP Header (8 bytes)