The TCP Protocol

The TCP Protocol

1. TCP Overview

TCP (Transmission Control Protocol) is the connection-oriented protocol of the transport layer and provides a reliable data transfer service.

Basic Characteristics of TCP

• Connection-oriented: a connection has to be established beforehand
• Reliable transmission: guarantees that data is delivered intact, in order, and without duplication
• Flow control: prevents the sender from going so fast that the receiver’s buffer overflows
• Congestion control: adapts to the state of the network and avoids congestion

Where TCP Applies

• File transfer: FTP, HTTP
• Web browsing: HTTPS
• Mail transfer: SMTP, POP3, IMAP
• Remote login: SSH, Telnet

2. The TCP Segment Structure

The TCP Header Format

Field Descriptions

Port number fields:

• Source port (16 bits): the port number of the sending process
• Destination port (16 bits): the port number of the receiving process

Sequence number fields:

• Sequence number (32 bits): the sequence number of the first byte of this segment
• Acknowledgment number (32 bits): the sequence number of the next byte the receiver expects

Control fields:

• Data offset (4 bits): the length of the TCP header, in units of 4 bytes
• Reserved (6 bits): reserved for future use, set to 0
• Flags (6 bits):
  • URG: the urgent pointer is valid
  • ACK: the acknowledgment number is valid
  • PSH: push flag
  • RST: reset the connection
  • SYN: synchronise sequence numbers
  • FIN: finish the connection

Window and checksum fields:

• Window size (16 bits): the size of the receive window
• Checksum (16 bits): the checksum over the header and the data
• Urgent pointer (16 bits): the last byte of the urgent data

3. TCP Connection Management

The Three-Way Handshake (Establishing a Connection)

The procedure:

1. SYN=1, seq=x: the client sends a connection request
2. SYN=1, ACK=1, seq=y, ack=x+1: the server acknowledges the connection
3. ACK=1, seq=x+1, ack=y+1: the client acknowledges the connection

State changes:

• Client: CLOSED → SYN_SENT → ESTABLISHED
• Server: CLOSED → LISTEN → SYN_RCVD → ESTABLISHED

The Four-Way Teardown (Releasing a Connection)

The procedure:

1. FIN=1, seq=u: the client sends a termination request
2. ACK=1, ack=u+1: the server acknowledges the termination request
3. FIN=1, seq=v: the server sends a termination request
4. ACK=1, ack=v+1: the client acknowledges the termination request

State changes:

• Client: ESTABLISHED → FIN_WAIT_1 → FIN_WAIT_2 → TIME_WAIT → CLOSED
• Server: ESTABLISHED → CLOSE_WAIT → LAST_ACK → CLOSED

4. Reliable Transmission in TCP

The Sequence Number Mechanism

• Byte numbering: every byte carries a unique sequence number
• Initial sequence number: chosen at random when the connection is established
• Sequence number wraparound: numbering restarts once the maximum is reached

The Acknowledgement Mechanism

• Cumulative acknowledgement: the acknowledgment number states the next byte expected
• Selective acknowledgement: acknowledges a particular segment
• Delayed acknowledgement: the acknowledgement is sent later, to reduce network overhead

Timeout and Retransmission

• Timeout: adjusted dynamically, based on the RTT
• Retransmission strategy: an exponential backoff algorithm
• Fast retransmit: retransmit immediately on three duplicate ACKs

The Sliding Window

• Send window: the range of data that may be sent but not yet acknowledged
• Receive window: the range of data that may be received
• Sliding: the window moves forward once an acknowledgement arrives

5. TCP Flow Control

The Purpose of Flow Control

• Preventing overload: stops the sender from going so fast that the receiver’s buffer overflows
• Matching rates: the sending rate matches the receiving rate
• Improving efficiency: makes full use of the receiver’s processing capacity

The Sliding Window Mechanism

• Window size: the size of the window advertised by the receiver
• Dynamic adjustment: adjusted to the receiver’s processing capacity
• Zero window: the receiver advertises a zero window when its buffer is full

Flow Control Algorithms

• Slow start: after the connection is established, start from 1 MSS
• Congestion avoidance: once the window reaches the threshold, grow it linearly
• Fast retransmit: retransmit immediately on three duplicate ACKs
• Fast recovery: after the retransmission, enter congestion avoidance

6. TCP Congestion Control

The Purpose of Congestion Control

• Avoiding congestion: prevents the network from becoming congested
• Fairness: several connections share the bandwidth fairly
• Efficiency: makes full use of the network’s resources

Congestion Control Algorithms

Slow start:

• The initial cwnd is 1 MSS
• cwnd doubles on every ACK received
• Until ssthresh is reached

Congestion avoidance:

• Once cwnd reaches ssthresh it grows linearly
• One MSS is added per RTT
• Until congestion occurs

Fast retransmit and fast recovery:

• Retransmit immediately on three duplicate ACKs
• After the retransmission, enter congestion avoidance
• Do not enter slow start

Congestion Control Parameters

• cwnd: the congestion window size
• ssthresh: the slow start threshold
• RTT: the round-trip time
• MSS: the maximum segment size

7. TCP Applications

Web Applications

• HTTP: the Hypertext Transfer Protocol
• HTTPS: HTTP over TLS
• Characteristic: reliable transmission, suitable for transferring documents

File Transfer

• FTP: the File Transfer Protocol
• SFTP: the SSH File Transfer Protocol
• Characteristic: large transfers, with a strong reliability requirement

Mail Transfer

• SMTP: the Simple Mail Transfer Protocol
• POP3: the Post Office Protocol
• IMAP: the Internet Message Access Protocol

Remote Login

• SSH: the Secure Shell protocol
• Telnet: the remote login protocol
• Characteristic: interactive communication, with a strong reliability requirement

Exercises

Exercise 1

How does TCP achieve reliable transmission?

Exercise 2

How do TCP flow control and TCP congestion control differ?

---

Glossary