Why are networks built in layers instead of as one monolithic system?

Layering decomposes a hard problem into smaller ones with clean interfaces, so each layer can be designed, implemented, and replaced independently. A new physical medium can be added without changing applications, and a new application can run over any network that provides the expected lower-layer services.

Is the Internet packet-switched or circuit-switched?

The Internet is fundamentally packet-switched: data is broken into packets that are forwarded hop by hop and share link capacity statistically. Some access and transport technologies emulate circuits or reserve resources, but the core IP layer uses packet switching.

Network Architecture and Layering

Network architecture is the organization of a communication system into layers of protocols, each providing services to the layer above and building on the layer below, so that the complex problem of moving data between hosts is decomposed into manageable, well-defined pieces.

Definition

A network architecture is a structured set of layers and protocols, together with the services each layer offers and the interfaces between layers, that together define how data is communicated across a network.

Scope

This area covers the structural principles of computer networks: the layered protocol stacks (the seven-layer OSI reference model and the five-layer Internet/TCP-IP model), the distinction between the network edge (hosts and access networks) and the network core (switches and routers), the two fundamental switching strategies of packet switching and circuit switching, and the metrics used to characterize network performance such as throughput, delay, and packet loss. It treats architecture as the framing context for all the lower- and higher-layer mechanisms covered in sibling areas, rather than the detailed mechanisms themselves.

Sub-topics

Core questions

Why is network functionality organized into layers, and what are the costs and benefits of layering?
How do the OSI seven-layer model and the Internet five-layer model differ, and why did the Internet model prevail in practice?
What distinguishes the network edge from the network core?
How do packet switching and circuit switching differ in resource sharing, delay, and reliability?
What metrics — throughput, latency, jitter, and loss — characterize the performance a network delivers?

Key concepts

protocol layers
OSI reference model
TCP/IP (Internet) model
encapsulation
network edge and core
packet switching
circuit switching
statistical multiplexing
throughput, delay, and loss
the end-to-end argument

Key theories

Protocol layering and the service-interface abstraction: Networks are organized as a stack of layers where each layer offers a defined service to the layer above through an interface, hiding the implementation below; this modularity lets layers evolve independently as long as interfaces are preserved.
The end-to-end argument: Functions such as reliable delivery and security are best implemented at the communicating end hosts rather than in the network core, because the core often cannot complete them correctly and would impose cost on flows that do not need them; this principle shaped the Internet's design.
Statistical multiplexing in packet switching: Packet switching shares link capacity on demand among many flows, achieving high utilization when traffic is bursty, at the cost of variable queueing delay and possible loss — in contrast to circuit switching's reserved but potentially idle capacity.

Clinical relevance

Layered architecture is the organizing principle behind essentially all modern networking: it lets a web browser, a Wi-Fi adapter, and an undersea fiber link interoperate without any one component knowing the internals of the others. Understanding edge-versus-core and packet-versus-circuit switching is foundational for capacity planning, cloud data-center design, and reasoning about latency-sensitive applications such as video conferencing and online gaming.

History

Layered communication architecture crystallized in the 1970s with the ARPANET and the parallel development of the OSI reference model, formalized by Zimmermann in 1980. The TCP/IP protocol suite of Cerf and Kahn, and the design philosophy articulated by Clark in 1988, established the simpler Internet model that ultimately dominated. The tension between the comprehensive OSI stack and the pragmatic TCP/IP stack defined much of networking's formative debate.

Debates

OSI seven-layer model versus the TCP/IP model: The OSI model is a richer pedagogical reference with distinct session and presentation layers, but the leaner five-layer TCP/IP model matched real Internet implementations and prevailed in practice; textbooks still teach both because OSI clarifies functions the Internet stack collapses.

Key figures

Hubert Zimmermann
David D. Clark
Vinton Cerf
Robert Kahn
Andrew S. Tanenbaum

Seminal works

kurose2021
tanenbaum2010
clark1988

Frequently asked questions

Why are networks built in layers instead of as one monolithic system?: Layering decomposes a hard problem into smaller ones with clean interfaces, so each layer can be designed, implemented, and replaced independently. A new physical medium can be added without changing applications, and a new application can run over any network that provides the expected lower-layer services.
Is the Internet packet-switched or circuit-switched?: The Internet is fundamentally packet-switched: data is broken into packets that are forwarded hop by hop and share link capacity statistically. Some access and transport technologies emulate circuits or reserve resources, but the core IP layer uses packet switching.