How the Internet Works: Network Edge, Packet Switching, Routing, and the TCP/IP Stack
A deep technical guide to how the Internet works, from end systems and access networks to packet switching, routing, DNS, HTTP, TCP/IP, CDNs, cloud backbones, and edge networks.
Free Claude Code Course with Lydia Hallie, Anthropic (Sponsor)
We partnered with Anthropic to make our Claude Code course free for everyone. No subscription, no trial. Just dive in.
It’s taught by Lydia Hallie, who’s been an instructor with us for years and now works on the Claude Code team at Anthropic. When she taught Claude Code live, it broke every platform record we have with over 10,000 people tuning in.
Lydia has a knack for visualizing how tools work under the hood, which is exactly the mental model you need to stop guessing with AI and start directing it.
The Internet is often described as a “network of networks,” and that phrase is still the best starting point. It is not one giant centralized system owned by a single company. It is a global collection of interconnected networks that communicate through standardized protocols. Cloudflare defines the Internet as a global network of interconnected computer networks that communicate using standard protocols, while the IETF’s original Internet Protocol specification describes IP as designed for interconnected packet-switched computer networks.
That design choice is what makes the Internet powerful. A phone on a mobile network in Manila can communicate with a server in Singapore, a CDN cache in Tokyo, a database replica in Oregon, or an API hosted on Google Cloud, AWS, or Azure because all these systems agree on common rules: IP addressing, routing, transport protocols, naming systems, and application protocols. The Internet works because independent networks can interoperate without being identical.
Internet Evolution
The Internet began as a research effort around packet-switched networking. Early ARPANET work showed that data could be split into packets and sent across shared infrastructure instead of requiring a dedicated circuit for every communication session. The transition from ARPANET’s earlier Network Control Protocol to TCP/IP in the early 1980s became a foundational moment in internetworking.
The World Wide Web came later. Tim Berners-Lee invented the Web at CERN in 1989, then wrote the first web client and server in 1990. W3C notes that his early specifications for URIs, HTTP, and HTML became the basis of the Web as it spread. This distinction matters: the Internet is the underlying network infrastructure; the Web is one major application built on top of it.
Since then, the Internet has evolved from academic and government networks into commercial broadband, mobile data, cloud computing, CDNs, streaming platforms, and edge infrastructure. Akamai describes CDNs as globally distributed edge servers that cache and deliver content closer to users to reduce latency, while Netflix’s Open Connect program localizes video traffic through ISP partnerships and embedded appliances.
1. The Network Edge
The network edge is where Internet communication starts and ends. It includes laptops, phones, desktops, smart TVs, IoT devices, enterprise workstations, cloud-hosted applications, and servers. In networking terms, these are often called end systems or hosts because they run applications such as browsers, email clients, web servers, APIs, streaming services, and mobile apps.
The modern edge is much broader than the old client/server model. A user’s phone may be a client when opening a website, but it can also upload video, stream live audio, sync files, run real-time messaging, or send telemetry from sensors. A server may live in a cloud region, a private enterprise data center, a CDN point of presence, or even inside an ISP network. Akamai describes edge servers as compute, storage, networking, and security resources placed closer to where data is produced and consumed.
This is why today’s Internet feels faster than a simple “user connects to distant server” model would suggest. Many requests no longer travel all the way to an origin server. They may terminate at a CDN, edge cache, reverse proxy, load balancer, security gateway, or regional cloud frontend. Cloudflare, Akamai, AWS, Google Cloud, Microsoft Azure, and Netflix all use some form of edge or backbone optimization to move traffic closer to users or keep it on high-performance private networks.
2. Access Networks
The access network is the part of the network that connects an end system to the first router or provider network on the path to the wider Internet. For a home user, this may be fiber, cable broadband, DSL, fixed wireless, Wi-Fi, or a home router connected to an ISP. For a business, it may be Ethernet, enterprise Wi-Fi, MPLS, SD-WAN, VPN, or a dedicated cloud interconnect. For mobile users, it is the cellular radio access network.
Home networks usually combine local connectivity and Internet access. Wi-Fi connects devices wirelessly to a router or access point, while Ethernet provides wired local connectivity. Cisco describes a wireless access point as a device that allows wireless-capable devices to connect to a wired network.
Institutional networks are more complex. Offices, schools, hospitals, factories, and government agencies often use switches, routers, access points, firewalls, identity systems, VPNs, VLANs, and monitoring tools. These networks do not merely “connect to the Internet”; they enforce access policies, segment traffic, prioritize applications, and secure internal systems.
Mobile access networks introduce another layer of complexity. Ericsson describes the 5G Radio Access Network as the link that connects wireless cellular devices to the core network, while Nokia describes the RAN as the final link between the network and the phone, including antennas, radios, and base stations.
In short, access networks are the Internet’s on-ramp. They decide the first part of the user experience: signal quality, bandwidth, congestion, latency, reliability, and sometimes whether traffic is routed efficiently at all.
3. Packet Switching at Global Scale
Once traffic leaves the access network, it enters the network core: a mesh of routers, backbone links, submarine cables, fiber routes, Internet exchange points, private cloud networks, and transit providers. The core does not send one continuous stream as a single object. It uses packet switching.
In packet switching, application data is broken into smaller units called packets or datagrams. Each packet carries addressing information that helps routers move it toward its destination. The IETF’s RFC 791 describes IP as transmitting blocks of data called datagrams from sources to destinations identified by addresses. Cloudflare similarly explains that data traversing the Internet is divided into packets, and IP information helps routers send those packets to the right place.
This design allows many users and applications to share the same infrastructure. Your video call, someone else’s banking transaction, another user’s Netflix stream, and a company’s API traffic may all share network links, but they are separated into packets and forwarded independently.
Packet switching is also why the Internet is resilient. Packets do not require one fixed end-to-end circuit. If routes change, congestion appears, or a link fails, routing systems can shift traffic to alternate paths. However, that flexibility also introduces complexity: packets may arrive out of order, experience delay, or get dropped. Higher-layer protocols such as TCP help applications recover from those conditions.
4. Forwarding vs. Routing
Routers perform two related but different jobs: forwarding and routing.
Forwarding is the local action. A packet arrives on one interface of a router. The router examines the destination IP address, checks its forwarding table, and sends the packet out through the appropriate next interface.
Routing is the broader control-plane process that decides what should be in those forwarding tables. Cisco describes routing as forwarding IP packets from one network to another, while Cloudflare explains that BGP is the routing protocol used between autonomous systems on the Internet.
The key nuance: routing does not simply mean “choose the shortest geographic path.” At Internet scale, routes are influenced by autonomous system relationships, peering agreements, routing policies, performance, failures, congestion, and business rules. Cloudflare notes that autonomous systems announce routing policy through BGP, and Google Cloud’s Cloud Router documentation shows how BGP is also used in cloud networking to dynamically exchange routes between VPCs and peer networks.
This is why traffic between two nearby locations can sometimes take a surprising path. The Internet is optimized through a mix of technical and commercial decisions, not just physical distance.
5. The Rules That Make Interoperability Possible
Protocols are the Internet’s shared language. They define message formats, timing, sequencing, addressing, error behavior, and expected responses. Without protocols, a phone, router, browser, server, and cloud load balancer would not know how to communicate.
DNS is one of the most important because humans use names, but networks route using addresses. Cloudflare describes DNS as the “phonebook of the Internet” that translates human-readable domain names to machine-readable IP addresses, and IBM describes DNS as the component of the Internet standard protocol responsible for converting domain names into IP addresses.
HTTP is another major protocol. It defines how browsers and servers exchange web requests and responses. MDN describes HTTP as a client-server protocol where clients send requests and servers return responses, while RFC 9110 defines HTTP semantics including status codes and request/response behavior.
TCP provides reliable, ordered delivery for applications that need correctness, such as web browsing, email, database connections, and file transfers. RFC 9293 defines TCP as providing a reliable, in-order byte-stream service to applications.
6. The TCP/IP Stack: A Layered Model for Internet Communication
The Internet Protocol stack is usually explained as four layers: application, transport, Internet/network, and link. This layered design keeps the Internet manageable. Each layer has a specific job and relies on the layer below it.
Application Layer
The application layer is where user-facing protocols live: HTTP for the Web, SMTP for email, DNS for name resolution, FTP for file transfer, and many API protocols used by modern apps.
When you open a website, your browser is not thinking about fiber cables or router forwarding tables. It creates an HTTP request. That request is passed down to the transport layer, then to IP, then to the local link layer.
Transport Layer
The transport layer manages communication between applications on different hosts. TCP focuses on reliability and ordering. UDP is lighter and faster but does not provide the same reliability guarantees by itself. Newer protocols such as QUIC build transport-layer behavior over UDP to reduce latency and improve performance for modern web applications.
TCP remains foundational because many applications need reliable byte streams. The IETF’s TCP specification describes TCP as conveying application byte streams through TCP segments carried inside IP datagrams.
Internet / Network Layer
The network layer is where IP lives. IP is responsible for addressing and routing packets across networks. IPv4 and IPv6 are the two major versions. IPv6 was standardized as the latest Internet Standard through RFC 8200, while IPv4 remains widely used across the public Internet.
IP is often called the “narrow waist” of the Internet because many applications can run above it, and many physical technologies can run below it. Whether the access link is fiber, Ethernet, Wi-Fi, 4G, 5G, or satellite, IP gives the Internet a common addressing and packet-delivery model.
Link Layer
The link layer handles communication on a local network segment. Ethernet, Wi-Fi, and cellular radio technologies all help move data across the immediate connection before packets are routed farther across the Internet. This layer deals with frames, MAC addresses, radio signals, cables, and local delivery.
7. What Happens When You Load a Website
A simple web page load involves many layers working together.
First, the user enters a domain name. DNS resolves that name into an IP address. Without DNS, users would need to remember numeric IP addresses instead of names.
Second, the browser establishes a connection to the destination service. For many web requests, this uses TCP and TLS; for HTTP/3, it may use QUIC over UDP. The transport layer prepares application data for transmission.
Third, the browser sends an HTTP request. The server, CDN, or edge proxy responds with HTML, JSON, images, scripts, video chunks, or other content. HTTP defines the structure of those requests and responses.
Fourth, the response is split into packets. IP addresses guide those packets across networks. Routers forward each packet hop by hop until it reaches the user’s device.
Finally, the receiving device reassembles the data and the browser renders the page. What feels like one click is actually a coordinated sequence of DNS resolution, connection setup, routing, packet forwarding, transport behavior, HTTP exchange, caching, and rendering.
8. CDNs, Anycast, Cloud Backbones, and Edge Compute
The original Internet model was host-to-host packet delivery. The modern Internet is still based on that model, but large-scale platforms now add layers of optimization.
CDNs cache content close to users. Akamai describes a CDN as a globally distributed network of edge servers that caches and delivers web content closer to users to reduce latency. Netflix’s Open Connect does something similar for streaming by localizing Netflix traffic through embedded appliances and settlement-free interconnection with ISPs.
Anycast lets the same IP address be announced from multiple locations. Cloudflare explains that anycast can route incoming traffic to the nearest data center with capacity to process the request, improving resilience against congestion and DDoS attacks. AWS Global Accelerator similarly uses anycast IP addresses from AWS edge locations so traffic can enter the AWS global network close to users.
Cloud providers also operate private global backbones. Google Cloud’s Premium Tier keeps traffic on Google’s backbone for much of the journey, while Microsoft says its global network connects Microsoft data centers and edge nodes around the world.
This means the “public Internet” experience is often shaped by private infrastructure. A request may enter a cloud edge location nearby, ride a private backbone across continents, hit a regional load balancer, and only then reach an application backend.
Why This Matters
For developers, SREs, platform engineers, and security teams, Internet architecture is not just theory. It directly affects application performance, reliability, cost, and risk.
Latency is shaped by physical distance, routing paths, DNS resolution, congestion, TLS handshakes, server placement, and cache hit rates. That is why CDNs, edge computing, anycast routing, and cloud backbone choices matter. Google Cloud’s global load balancing, for example, uses a single anycast IP address as a frontend for backend instances across regions, with failover when backends become unhealthy.
Reliability depends on redundancy across layers. A resilient system needs more than multiple servers. It needs DNS strategy, regional failover, multiple network paths, healthy routing, observability, and protection from provider-level incidents.
Security also starts at the network layer. DNS abuse, DDoS attacks, BGP route leaks, misconfigured firewalls, exposed services, and weak segmentation can all break applications before code-level defenses even matter. Cloudflare notes that BGP routing information is essential for large-scale Internet operation, but BGP is also a trust-heavy system that requires careful validation and monitoring.
Summary
The Internet works because independent networks agree on common protocols. End systems at the edge generate and consume data. Access networks connect users and institutions to service providers. The network core uses packet switching to move data across routers and backbone links. Routing systems decide where packets should go; forwarding moves them hop by hop. DNS turns names into addresses. TCP, UDP, HTTP, IP, Ethernet, Wi-Fi, and other protocols each handle different parts of the communication process.
The TCP/IP stack gives this complexity structure. The application layer defines user-facing behavior, the transport layer manages end-to-end delivery, the network layer handles addressing and routing, and the link layer moves data across local media.
The modern Internet is now heavily optimized by CDNs, cloud backbones, anycast networks, mobile RANs, and edge infrastructure. But underneath all of that, the same core idea remains: break data into packets, address them, route them across interconnected networks, and reassemble them at the destination. That simple architectural principle is what allowed the Internet to evolve from a research network into the foundation of modern communication, commerce, cloud computing, streaming, mobile apps, and AI-era digital infrastructure.
The AI notepad for people in back-to-back meetings (Sponsor)
Most AI note-takers just transcribe what was said and send you a summary after the call.
Granola is an AI notepad. And that difference matters.
You start with a clean, simple notepad. You jot down what matters to youand, in the background, Granola transcribes the meeting.
When the meeting ends, Granola uses your notes to generate clearer summaries, action items, and next steps, all from your point of view.
Then comes the powerful part: you can chat with your notes. Use Recipes (pre-made prompts) to write follow-up emails, pull out decisions, prep for your next meeting, or turn conversations into real work in seconds.
Think of it as a super-smart notes app that actually understands your meetings.
Free 1 month with the code SCOOP