OSI vs TCP/IP

OSI is a reference model from the ISO in the 1980s with seven layers; TCP/IP is the protocol suite the internet runs, usually described with four or five layers (link, internet/IP, transport, application — with physical sometimes split out). The internet protocols were designed and deployed before OSI was finished, so the seven layers never described them exactly: there is no separate session or presentation protocol in the stack; TLS sits awkwardly between transport and application, and HTTP does its own "presentation" (content types, encodings).

What both models get right is encapsulation: each layer adds a header, treats everything above it as opaque payload, and offers a service to the layer above without exposing how. IP moves packets between hosts without caring whether TCP or UDP is inside; TCP gives an ordered byte stream without caring whether HTTP or SSH is inside. That independence is why a new application protocol needs no router changes, and why routers can be replaced without applications noticing.

In practice the numbers survive as vocabulary: an "L2 switch" forwards by MAC, an "L3 router" by IP, an "L4 load balancer" by IP and port, an "L7 proxy" by HTTP content. Engineers say "layer 7" and "layer 4" daily and never say "layer 5 or 6". So the difference matters as language, not as architecture — and the layer boundaries are where you debug: which layer produced this failure, and what tool observes that layer?