The Upper Layers: Application, Presentation, and Session
One thing that has always been clear about the OSI model is that the Application layer includes application data and application protocols. The Hypertext Transfer Protocol (HTTP) is an application protocol that a web browser uses for communicating with web servers. Application data would be an HTTP GET request that the browser sends to a web server. Likewise, the web page that the server sends in response would also be application data. In short, application data is whatever the application sends or receives over the network.
Incidentally, an application can use more than one protocol. For example, when a web browser uses the Hypertext Transfer Protocol Secure (HTTPS) protocol to send a request to a web server, it's making use of two protocols: HTTP and Transport Layer Security (TLS). Despite the latter's confusing name, both are application protocols.
For all practical purposes, the upper layers (Session, Presentation, and Application) are one layer: the Application layer. The actual functions of the Session and Presentation layers—things like authentication and negotiating an application protocol—occur in the application anyway. They don't include any network functions and are concerned only with application data and application protocols.
Making Sense of Layers
The ISO never clearly defined what a layer is. The closest they came was a circular definition. But we can infer from the OSI reference model what they had in mind.
The concept of layering comes straight from software development (many of the OSI folks were operating system developers). The idea was that applications would treat the network as a software abstraction, somewhat like a filesystem. A filesystem acts as a layer that sits between the application and physical storage (e.g., disks). When the application needs to store some data, it just sends that data to the filesystem layer, which in turn takes care of the specifics of writing it to disk.
The OSI folks thought that in the same way that an application can store data on a filesystem without having to know anything about the underlying disks, so could it also send data over a network without requiring any network-specific coding or knowing anything about the network's infrastructure. Each layer would consist of a set of network-related functions implemented by the operating system or some middleware that would sit between the application and the host's physical network interface. Collectively, these layers would handle all the mechanisms of getting the application data onto the network and giving the network enough information to make sure the data got to its destination.
With the exception of the Physical layer, the layers of the OSI model are purely imaginary. Just as a filesystem is a software abstraction that hides the details of physical storage, the layers of the OSI model are just collections of software functions that hide the details of the network from applications and users. You can't see a filesystem with your eyes in the same way that you can see a hard drive, and you can't see the Data Link layer in the same way that you can see a switch. Layers are software abstractions and nothing more.
Figure 1.1 illustrates the concept of how layering might work using the Transmission Control Protocol (TCP) and Internet Protocol (IP), which are both included in the kernels of modern operating systems (Linux, Unix, and Windows). Keep in mind that the only real objects in this figure are the host and the physical network interface.
Figure 1.1 How layers abstract the network from an application
In this high-level example, when an application needs to send data it places the data in what the OSI model generically calls an application protocol data unit (PDU). The specifics of the application PDU aren't important and, with the exception of firewalls that do deep packet inspection, are opaque to the network. The application passes its PDU to a protocol in the layer directly below, as shown in Figure 1.2. The protocol generates a new PDU and tacks the application PDU onto the end of it—a process called encapsulation. It then passes this new PDU down to a protocol at the next lower layer, and so on. What ends up on the wire is a giant PDU that contains several smaller PDUs from the protocols operating at the higher layers. Later in the chapter we'll walk through a detailed example of how encapsulation works, but first, we need to talk about what happens at each of these lower layers.
Figure 1.2 At each layer, data is encapsulated in a PDU and passed down to the next lower layer.
The Lower Layers: Physical, Data Link, Network, and Transport
The purpose of a network is to allow applications running on different hosts to communicate with one another. Robert Metcalfe, one of the inventors of the original Ethernet, said it succinctly in 1972: “Networking is interprocess communication.” Thus, at a minimum, a network needs to perform three basic functions:
Layer 1: Physical Connectivity between Nodes A node can be a workstation, server, router, switch, firewall, or any network-connected device that has a processor and memory.
Layer 2: Node-to-Node Data Transfer Data transfer between two nodes physically connected to a shared medium.
Layer 3: Forwarding/Routing Data transfer between any two nodes, regardless of whether they're physically connected to the same medium.
The OSI model sorts these three functions along with many others into the first four layers of the OSI model, as shown in Table 1.2. Not all protocols that operate in a given layer implement all the functions listed for that layer.
