The OSI Seven-Layer Reference Model and the Internet Protocol Model

The OSI Model - A Surviving Remnant of the OSI Protocol Stack

I remember that way back in the 1980s, when I was at Bell Labs, even in the days before the divestiture, there was a lot of excitement about OSI – Open Systems Interconnection (though everyone I knew, including me, called it “Open Systems Interconnect”). OSI was meant to be the protocol stack to replace all proprietary protocol stacks – the answer to the proliferation of proprietary protocols (how’s that for alliteration?). The popular data communications periodicals of the day, such as Data Communications, had numerous articles telling us of the great days ahead in which we would be freed from being enslaved to either IBM with SNA or Digital with DECnet.

In fact, those great days did arrive, but as we all know, it is TCP/IP that rescued us and not OSI. But that is not what the pundits thought back then. Here is the ending paragraph from an article in the April 22, 1986 issue of Computer Decisions, entitled "TCP/IP: Stairway to OSI" by Robert A. Moskowitz:

"Over the long haul, most vendors are going to migrate from TCP/IP to support layer 4, the transport layer, of the OSI model. For the short term, however, TCP/IP provides organizations with enough functionality to protect their existing equipment investment. And over the long term, TCP/IP promises to allow for easy migration to OSI."

That will teach you to believe everything you read!

It is not my intent to discuss the reasons for the demise of OSI and the concomitant success of TCP/IP – you can go to Wikipedia and see a discussion of OSI and some reactions, as well. It really doesn’t matter to most of us these days.

But there is one reason that I still mention OSI at all in my communications protocols class: we inherited some of the terminology that was popularized by the OSI documentation, words such as “end-system” (which is a “host” in the IP world) and “intermediate-system”, and we inherited the OSI seven-layer model, a model of protocol layers that became a standard in how communications protocols are taught, in spite of the fact that it does not realistically represent the protocols with which we internetwork. The functions of the seven layers may very well be present in many communications environments, but it might be a stack of four or five protocols that accomplishes those functions. However, I have not seen the textbook on computer networking today that does not pay at least some lip service to the seven-layer model, so it seems to have become some kind of obeisance that we must pay.

Influence on the Internet Protocol Model

The protocol stacks that are presented as the Internet protocol stack also show influence of the OSI model – much of the terminology is the same, and textbooks typically present an Internet protocol stack that uses OSI terminology and that includes a physical and data link layer. This wasn't always the case. In a 1981 article (Computer Networks 5(1981), pp. 261-271) by Jon Postel, Carl Sunshine, and Danny Cohen there is a diagram showing the Internet Protocol functioning on top of one of several Network Protocols, and the text describes such a network protocols as being "the packet transmission protocol in each individual network." In a 1983 article (Computer Networks 7(1983), pp. 307-318) by Vinton Cerf - often referred to as "one of the founding fathers of the Internet" - and Edward Cain we find (Figure 4) that the layer that IP functions at is called the "Internetwork layer" and the layer under that is called the "Network layer" - not at all consistent with the common terminology today. Today, we universally find that the layer at which IP functions is called the "Network Layer", and the very important layer that was called the "Network Layer" is usually omitted.

The Internet Protocol Model

In fact, from an Internet perspective, it makes a lot more sense to include a layer under IP (such a layer under IP once being called the "Network Layer") but above the infrastructure (typically made up of what are called the data link and physical layers) and to entitle this intermediary layer the "Network Interface Layer" – a layer that defines how IP operates over a particular infrastructure.

Under the Network Interface Layer would be the infrastructure layer, which is not defined at all by the IETF, but rather by standards organizations that define infrastructure protocols. The infrastructure layer might in fact have many layers of its own – not just physical and data link layers. Take X.25 (a real ancient infrastructure protocol) – it has three layers. Take ATM (a functional descendent of X.25) – it, too, has more than two layers, and its layers are divided into sublayers. Or, take a look at how IP is typically implemented over ADSL and see why the traditional "data link layer and physical layer" model doesn't make any sense. That’s why I prefer to present the following layered structure for the Internet protocols:

The Network Interface Layer

With this model, the Network Interface Layer would then be defined according to the relevant RFC of the type "IP over ...", such as RFC 894 "Standard for the Transmission of IP Datagrams over Ethernet Networks" or RFC 2549 "IP over Avian Carriers with QoS". It is with this model in mind that the drop-down menus on the RAD University website have been organized.

When reading the various tutorials on RAD University, it is likely that you will see much reference to the OSI protocol stack, but bear in mind that it is this IP stack that is more representative of the protocols in use today.

The Rest of the Layers in the Internet Model

I'll just briefly summarize the purpose of the other layers.

The application layer is where any of the application protocols, such as SMTP for mail or FTP for file transfer, function and interface to a transport layer protocol. A transport layer protocol is an end-to-end protocol, and the transport layer protocols in the Internet suite of protocols include TCP and UDP, TCP being connection-oriented and reliable and UDP being connectionless and unreliable (meaning no acknowledgment of received messages is sent by UDP).

And then, of course, there is IP, the protocol for connectionless datagram delivery, which, because of the influence of the OSI Reference Model is referred to as the Network Layer.

The reader is welcome to find tutorials about the protocols (hopefully more user-friendly and tutorial in nature than the RFCs) at all the layers by using the drop-down menus on the main tutorial page.

Debby Koren, "Dean" RAD University