The OSI model is perhaps the best-known—and perhaps the most-loved—model in the networking world. It’s taught in every basic networking course, and just about every blog (other than this one) has some article explaining the model someplace or another (for instance, here is one of the better examples).
The reality is, however, that I’ve been in the networking business for 30’ish years and I’ve never once used the OSI model for anything practical. I’ve used the model when writing books because just about every book on networking has to have a section on the OSI model. I’ve used the model when writing a paper comparing two different protocols, back in the multiprotocol days (VIP versus IPX versus IP), but we don’t have those kinds of arguments very often any longer.
So, we all learn the OSI model, and yet I don’t know of anyone who actually uses the OSI model in understanding how protocols work, or how to troubleshoot a network. There’s the “it’s a layer two problem” statement, and that’s about the end its useful life, it seems.
Let me make a suggestion—learn, use, and teach the RINA model instead. Okay, so what is the RINA model? It is the Recursive Internet Network Architecture, hence RINA. The observation John Day made when creating the RINA model is this: there are only four problems you need to solve to move data from one host to another. It just happens that those four problems need to be solved multiple times: across each physical link, between each pair of hosts, and then again between each pair of applications (at least—John Day argues there should be eight layers rather than seven, but that argument is outside the scope of this simple blog post).
The four problems are: transport (which I would call marshaling), multiplexing, error control, and flow control. Protocols that solve these problems tend to solve two of the four problems rather than one; the pairings almost always seem to be marshaling with multiplexing and error control with flow control.
On a single Ethernet hop, multiplexing and marshaling are generally solved by the physical link Ethernet protocol. Error and flow control, on the other hand, are generally solved by the data link protocol (something we don’t much think about any longer). Again, multiplexing and marshaling are generally solved by IP, while error and flow control are generally solved by TCP.
The model doesn’t perfectly describe the purpose of every protocol in the modern networking stack, but it comes much closer than the OSI model. Since it’s recursive, you can even insert layers to represent tunnels without a lot of hand-waving in the process, or “breaking” the model.
Why is the RINA model better for understanding protocols? Because it describes what the protocol is trying to accomplish, rather than where it lives in the stack, or what kinds of interfaces it provides. Why is the RINA model better for troubleshooting? Because if you know what isn’t working, it gives you a better general area to look in to find the problem.
My challenge for 2021 is, then—learn and use the RINA model. I (humbly) believe it will actually make you a better network engineer.