TCP

TCP (Transmission Control Protocol) provides a reliable end-to-end byte stream over an unreliable internetwork. Originally defined in 1981, there have been many amendments since [1, P. 552].

Table of contents

1. Introduction
2. The TCP service model
3. The TCP Protocol
   1. Segment header
4. TCP Connection Establishment
5. TCP connection release
6. TCP connection management modeling
7. TCP sliding window
8. TCP timer management
9. TCP congestion control
10. References

Introduction

A machine that supports TCP has a TCP transport entity, which is normally part of the kernel. The entity accepts data streams from local processes. It breaks the streams into chunks no larger than 64KB (usually 1460 bytes to fit in an Ethernet frame), and sends each chunk as an IP datagram. When datagrams containing TCP data arrive at a machine, they are given to a TCP entity that reconstructs the byte stream [1, P. 553].

IP doesn’t guarantee that datagrams will arrive, or how fast they are sent. So TCP is responsible for sending datagrams fast enough to maximize capacity, but not too fast to cause congestion. TCP must also reassemble messages in the correct order, in case they arrive out of order [1, P. 553].

The TCP service model

TCP service is achieved with endpoints called sockets. A socket has a socket number, which contains the IP address of the host and a 16-bit port number.

For TCP service to be obtained, a connection must be established between a socket on one machine and a socket on another. Sockets can handle multiple connections at a time.

Port numbers below 1024 are reserved for standard services. They are called well-known ports. Ports from 1024 to 49151 can be registered for use, but applications can choose there own ports without registering them.

TCP connections are full duplex and point-to-point. TCP doesn’t support multicasting or broadcasting [1, P. 554].

“A TCP connection is a byte stream, not a message stream. Message boundaries are not preserved end to end.” If a process writes four times to a TCP stream in 512-byte chunks, the data could be delivered as four 512-byte chunks, or two 1024-byte chunks. There’s no way for the receiver to detect the units the data was written in [1, P. 554].

TCP can choose to buffer the data it receives from a user process before sending it. This is to send a larger amount of data at once. Sometimes this behavior is undesirable. To force data out as soon as its received, TCP has a push flag that’s carried on packets, this can be set from user space by providing an OS flag check (like TCP_NODELAY in Linux) [1, P. 555].

The TCP Protocol

Every byte on a TCP connection has its own 32-bit sequence number. The sequence numbers are carried for sliding window in one direction and for acknowledgements in the other [1, P. 556].

TCP sends information in TCP segments. A segment contains a fixed 20-byte header (with an optional extra part) followed by 0 or more bytes of data. TCP software decides how big segments should be. The maximum segment size is 65,515 bytes (the max IP payload). Each link in a network has an MTU (Maximum Transfer Unit). Generally the MTU is 1,500 bytes (the Ethernet payload size), so 1,500 bytes is often the upper bound on the segment size.

It’s still possible for an IP packet to be fragmented when passing over a network where a link has a small MTU. To avoid this, modern TCP implementations implement path MTU discovery. This uses ICMP error messages to find the smallest segment sizes over the network [1, P. 556].

“TCP implements a sliding window protocol with a dynamic window size”. When a sender transmit a segment, it starts a timer. When the segment arrives at the destination, the receiving TCP entity sends back a segment (with data if any exist, and otherwise without) bearing an acknowledgement number equal to the next sequence number it expects to receive and the remaining window size. If the sender’s timer goes off before the acknowledgement is received, the sender transmits the segment again” [1, P. 556].

Segment header

    0                   1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |          Source Port          |       Destination Port        |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                        Sequence Number                        |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                    Acknowledgment Number                      |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |  Data |  Res- |C|E|U|A|P|R|S|F|                               |
   | Offset| erved |W|C|R|C|S|S|Y|I|            Window             |
   |       |       |R|E|G|K|H|T|N|N|                               |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |           Checksum            |         Urgent Pointer        |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                    Options                    |    Padding    |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                             Data                              |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

[2, Pp. 14 computer-networks]

The Source Port and Destination Port fields specify the TCP ports of the connection endpoints. A TCP port and an IP address identify a connection.

The Sequence Number field is the sequence number of the first data octet in the segment, except when SYN is present. “If SYN is present the sequence number is the initial sequence number (ISN) and the first data octet is ISN+1” [2, P. 16].

If the ACK bit is set, the Acknowledgement Number field contains “the value of the next sequence number the sender of the segment is expecting to receive” [2, P. 16].

The Data Offset field is the number of 32-bit words in the header, used to calculate where the data begins.

The control bits:

• CWR: Congestion Window Reduced field.
• ECE: Used to echo back the congestion indication.
• URG: Urgent Pointer field.
• ACK: Acknowledgment field.
• PSH: Push Function.
• RST: Reset the connection.
• SYN: Synchronize sequence numbers.
• FIN: No more data from sender.

[2, P. 16]

The Window field is “the number of octets beginning with the one indicated in the acknowledgment field which the sender of this segment is willing to accept”. The windowing mechanism is explained in the TCP Sliding Window section.

Checksum contains a checksum that’s used to improve reliability.

The Options field can contain extra options that aren’t covered by the header, for example the MSS (Maximum Segment Size) option. Another option is SACK (Selective ACKnowledgement), which is covered later in the notes [1, Pp. 559-60].

TCP Connection Establishment

TCP connections are established in a three-way-handshake.

One server must be listening for connections. The server listens for incoming connections by executing the LISTEN and ACCEPT primitives [1, P. 560].

To establish a connection, the client executes a CONNECT primitive with the IP address and port number of the server, and the maximum segment size it will accept. The CONNECT primitive sends a TCP segment with the SYN bit on and the ACK bit off and waits for a response [1, P. 560].

When the segment arrives at the server, the TCP entity checks to see if there is a process that is listening on the port. If there isn’t, it sends a response with the RST bit on to reject the connection [1, P. 560].

If a process is listening on the port, the process is given the incoming TCP segment. “It can either accept or reject the connection. If it accepts, an acknowledgement segment is sent back” [1, P. 560].

TCP connection release

Each TCP connection is released independently. To release a connection, either side of the connection can send a TCP segment with the FIN bit set. When the FIN bit is acknowledged, the direction is closed down for new data. “When both directions Have been shut down, the connection is released” [1, P. 562].

If there is no response to a FIN within two maximum packet lifetimes, the FIN sender releases the connection. The other side will eventually notice that nobody is listening, and will time out [1, P. 562].

TCP connection management modeling

The steps of a establishing and releasing a TCP connection can be represented as a finite state machine with 11 states.

[1, P. 563]

TCP sliding window

TCP sliding window is a method to control how much data is sent by a sender to a receiver. It’s used to avoid situations where data is sent too quickly to be processed by the receiver [2, P. 4].

The Window value is the number of octets that can be sent since the last acknowledged segment.

When the Window value is 0, the sender can’t send any data. There are two exceptions:

1. Data can be sent if it’s urgent.
2. A 1-byte segment can be sent to “force the receiver to reannounce the next byte expected and the window size” (a technique known as TCP window probe) [1, P. 566].

TCP timer management

TCP uses timers to do part of its work. These include:

• RTO timer
• Persistence timer

The RTO (Retransmission TimeOut) timer is used for retransmission. Each time a segment is sent, a retransmission timer is started. If the timer completes before the segment has been acknowledged, the segment is retransmitted. The timer value is calculated dynamically, and adjusted according to current network conditions [1, Pp. 568-9].

The persistence timer is used to avoid deadlock when a receiver sends an acknowledgement with a window of 0, but when the receiver updates the window, the message is lost. When the persistance timer runs, the sender transmits a window probe to ensure it has up-to-date window information [1, P. 571].

TCP congestion control

TCP is the main network protocol in the Internet stack that’s responsible for avoiding network congestion (IP simply drops packets).

TCP uses the window mechanism to control the data flow, and packet loss as the signal that there’s congestion. It maintains a congestion window, separate from the current window value. The congestion window size is the number of bytes that the sender can have in flight at a time. TCP stops sending traffic if either the congestion window or the window are full [1, Pp. 571-2].

The initial congestion window is set to a small value of at most 4 segments. Each time an acknowledgement is received, the sender sends 1 extra segments worth of bytes. This algorithm is called slow start, and it exponentially increases the amount of data sent [1, P. 574].

The sender keeps a slow start threshold for the connection. This is set high initially, so that it doesn’t limit the connection. When packet loss is detected, the slow start threshold is set to half the congestion window. When the slow start threshold is reached, TCP switches to additive increase. The congestion window is increased by 1 segment each round trip time [1, P. 576].

Instead of waiting for the retransmission timeout to detect packet loss, TCP also uses out-of-order packets as a signal that packets have been lost. When a receiver receives out of order packets, it sends duplicate acknowledgements to the sender so that the sender can retransmit the packet, which is presumed to be lost. This heuristic is known as fast retransmission [1, P. 577].

References

1. [1] A. Tanenbaum and D. Wetherall, Computer Networks, 5th ed. 2011.
2. [2] “Transmission Control Protocol,” no. 793. RFC Editor, Sep-1981.