OSI Model: What It Is, the 7 Layers, and How It Compares to TCP/IP

According to IBM, the OSI model (Open Systems Interconnection) is a conceptual model that organizes network communication into seven abstract layers, stacked one above the other, from the cable up to the program a user sees on screen. It isn’t software you install, and it isn’t a bundle of cables to wire up — it’s a way of thinking about the problem, built to separate the responsibilities of the people who design network hardware from the people who design network software. Readers of How the internet works: the journey of data from cable to page have already met pieces of this story without the name that ties them together.

The seven layers, read from top to bottom, are Application, Presentation, Session, Transport, Network, Data Link, and Physical, according to IBM. At the seventh, Application, live the protocols — HTTP among them — that turn network data into what a user sees when a browser requests a page. The sixth, Presentation, prepares that data by translating, compressing, and encrypting it before it reaches Application. The fifth, Session, opens and closes the connections between the two devices talking to each other.

Moving down the stack changes the kind of job each layer does. The Transport layer (4) delivers data end to end between two devices, checking that none of it is lost along the way. The Network layer (3) addresses and routes that data between devices sitting on different networks. The Data Link layer (2) instead handles the closer exchange, between two devices on the same network. At the base sits the Physical layer (1): the cables and routers that turn data into sequences of 1s and 0s, the same alphabet described in Binary Code: What It Is and How Computers Write Numbers.

The OSI stack, another name for the same ladder

“OSI stack” is another way people refer to this same model. The word stack describes layers piled one on top of the other.

The TCP/IP model, the one networks actually run on

Unlike the OSI model, TCP/IP relies on standardized protocols that are genuinely implemented in real networks, and it’s built from four layers instead of seven, according to IBM. RFC 1122, the IETF document that describes the requirements for Internet hosts, lists these four layers: Application, the top layer of the Internet protocol suite; Transport, which provides end-to-end communication services to applications; Internet, crossed by every transport protocol to move data from the source host to the destination host using the IP protocol; and the link layer, which a host must implement to communicate on the network it’s directly attached to. Cisco Networking Academy material describes the same four layers with different wording, useful for anyone who prefers another angle on the same facts: the Application layer represents data for the user and handles encoding and dialog control, the Transport layer supports communication between different devices on different networks, the Internet layer determines the best path across the network, and the Network Access layer controls the hardware devices and transmission media of the network.

Where the two models meet

Neither model is more “correct” than the other: they describe the same network communication at a different level of detail. According to Cisco Networking Academy, OSI’s Network layer (3) maps directly to TCP/IP’s Internet layer, and both are responsible for routing packets between different networks. In the same way, OSI’s Transport layer (4) maps to TCP/IP’s Transport layer.

The match gets messier higher up the stack. TCP/IP’s Application layer alone groups OSI layers 5, 6, and 7 — Session, Presentation, and Application — which remain a useful reference for people building application software but have no separate counterpart in TCP/IP. At the bottom of the stack, OSI keeps Data Link and Physical apart, which is why it’s usually the model people reach for when discussing the lowest layers of a network, the same ground covered in History of Computers: A Timeline from Babbage to the Web.

Protocols, layer by layer

Every layer has its own typical protocols. According to IBM, the Application layer runs HTTP, FTP, DNS, and SMTP; the Transport layer runs TCP and UDP; the Network layer runs IPv4 and IPv6; the Data Link layer runs Ethernet and ARP.

Practical example: when a browser loads a web page, HTTP (Application layer in OSI, Application in TCP/IP) puts the request together; TCP (Transport layer in both models) breaks it into packets and tracks their delivery; IPv4 or IPv6 (Network layer in OSI, Internet in TCP/IP) find the path between networks; and finally Ethernet (Data Link layer in OSI, Link in TCP/IP) carries the data over the last stretch, from the network card to the cable or the Wi-Fi antenna.

This breakdown into layers is what lets two computers running different hardware and software understand each other: each one speaks a different protocol at every layer, but the sequence of layers is shared, as also told in The History of the Internet: From ARPANET to the World Wide Web.