The Transport Layer: From One Host to One Program

The transport layer, layer 4, delivers data from a program on one machine to a program on another. The network layer stops at the machine: an IP address names a host, not the web server, database or mail server running on it. The transport layer adds a port number, a 16-bit number that names one program's endpoint on that host, so the operating system knows which program each arriving packet is for.

Picture a server with one IP address, 203.0.113.10, running a web server, an SSH server for remote logins and a PostgreSQL database. Packets for all three arrive at the same address. The port number in each one decides where it goes.

Think of the IP address as a building and the port as a flat number inside it: the postal service gets a letter to the building, and the flat number gets it to one household. The comparison breaks in one place. A flat holds one household, but one server port, such as 443, can hold thousands of connections at once, which the next ideas explain.

A program claims a port by listening on it: it asks the operating system to hand it everything that arrives for that number. Only one program can listen on a given port and transport at a time, which is why starting a second web server on port 443 fails with "address already in use".

The internet has two transports in wide use, and they differ in one question: what happens when a packet is lost. TCP (Transmission Control Protocol) notices, sends it again and delivers everything in order, so the program receives a complete, ordered stream. UDP (User Datagram Protocol) does nothing; each packet arrives or it does not, and the program deals with the gap.

Ports belong to a transport, so TCP port 53 and UDP port 53 are different endpoints, and a program can listen on both.

So when data reaches the right machine but the wrong program, or no program, the question is at layer 4: which port, and which transport.