Transport Layer

The Transport layer is the key to how networked applications talk to each other.

Important concept to grasp are:

  • Multiplexing, demultiplexing, and the use of network ports
  • How ports works with IP addresses (conceptualy, a network sockets)
  • Socket object (at implementation level)
  • Network (un)reliability and its engineering
  • Connection-oriented and connectionless protocol
  • Difference and trade-offs of TCP and UDP
  • Three-way-handshake (SYN, SYN ACK, ACK)
  • Congestion avoidance in TCP

Internet Protocol and its system of addressing intend to provide communication between hosts (devices). Hosts can be on the same or a different network and thanks to IP the communication works. However IP only permits a message from one host to the other and no more than that.

Because of this limitation, with IP alone we cannot communicate between different application running on different or even the same host.

Indeed there are generally multiple applications running on the same host. These applications all want to send and receive data simultaneously.

To do so, we need multiplexing and demultiplexing.

Multiplexing and Demultiplexing

Multiplexing, and its reverse process demultiplexing, as a general concept can be applied to a lots of contexts in communication network. Think how an optical fibres can carry multiple light signals at differnt angles of recration, or radio waves carrying signals at different frequencies. These are examples of multiplexing. At the Transport level though, multiplexing is done using network ports.

Ports

A port is like an integer between the range 0-65535 which serves as an identifier for a specific process running on a host. The range is separated in three sub-range:

  • 0-1023: well-known ports. Used for commonly used network services such as HTTP (80), HTTPS (443), SMTP (25), DNS (53) etc.
  • 1024-49151: registered ports. Assigned to private entities but sometimes used for ephemeral ports
  • 49152-65535: dynamic ports (or private ports). Used for customized services or for allocation as ephemeral ports.

Services running on a server directly are most likely using a port in the well-known range (first range). For example a web server using HTTP will likely have port 80 assigned to it. We say the web server is listening on port 80.

A service running on a client machine like in a browser, however, will instead use an ephemeral or temporary port assigned to it by the operating system (third range).

The source and destination port numbers are included in the Protocol Data Units (PDU) for the transport layer in two header fields aptly named “Source Port” and “Destination Port”.

Using both the IP address and the port number permets end-to-end communication between specific applications on a different machines. This combination is referred to as communication end-point. The communication end-point is referred to as a socket.

We can use the netstat -ntup command to produce a list of active network connections.

Sockets

The term “sockets” is used differently depending on the context. As a concept, it defines an endpoint for inter-process communication.

For Internet communication, it is as described above a inter-process communication mechanism between networked processes on different machine, but sometimes on the same.

UNIX sockets are a different beast: it’s a mechanism for communication between local processes running on the same machine.

In programming, a socket object is often instantiated; this object will take care of some of the network and communication-related tasks.

Most language follow the Berkley sockets API model for their implementation.

Protocol