How Does the Internet Actually Work? From Your Click to a Server and Back

When you type a URL and hit enter, your click triggers a cascade of coordinated systems across the world — DNS queries, routing decisions, packet transmission through undersea fibre-optic cables, server responses — all orchestrated in milliseconds by protocols designed in the 1970s. The internet is not a cloud but a tangible global infrastructure of millions of routers, datacentres, and cables. Remarkably, no single entity owns or controls it. This guide explains exactly what happens in those milliseconds when you click a link.

Step 1: You type a URL

You enter scihub101.com in your browser. What's actually happening:

The browser doesn't know what scihub101.com is. It needs an IP address — a unique numerical identifier like 76.76.21.21. To find it, it needs to make a DNS (Domain Name System) lookup.

Step 2: DNS — the internet's phonebook

DNS converts human-readable domain names into IP addresses. The lookup chain:

Recursive resolver

Your computer asks its configured DNS resolver — usually your ISP's, or a public one like Google's 8.8.8.8 or Cloudflare's 1.1.1.1. If the resolver has the answer cached, it returns it immediately.

Root servers

If not cached, the resolver asks one of 13 root server clusters scattered globally. These don't know the answer either — they tell the resolver where to find the .com servers.

TLD servers

The resolver asks the .com Top-Level Domain servers. They direct it to the authoritative nameservers for scihub101.com (typically managed by the domain owner's hosting or DNS provider).

Authoritative servers

These have the actual record. They return the IP address: "scihub101.com is at 76.76.21.21."

The resolver caches this answer and passes it back to your browser. Total time for a fresh DNS lookup: ~50-100 milliseconds.

Step 3: Establishing the connection

Your computer now has the IP address. It needs to establish a connection — usually using TCP (Transmission Control Protocol).

The TCP three-way handshake

1. Your computer sends a SYN packet: "Hi, want to talk?"
2. The server replies with SYN-ACK: "Sure, ready when you are."
3. Your computer sends ACK: "Got it, here we go."

These three packets establish a reliable connection. If any get lost, they're retransmitted.

TLS — adding encryption

For HTTPS sites (almost all of them now), an additional handshake happens after TCP:

• Browser and server agree on encryption keys.
• Server presents its certificate (issued by a trusted authority like Let's Encrypt).
• Both sides verify the certificate is valid and haven't been tampered with.

This entire setup typically takes another 100-200ms.

Step 4: HTTP request

Now the browser sends an HTTP request:

GET / HTTP/1.1
Host: scihub101.com
User-Agent: Mozilla/5.0...
Accept: text/html,...

This request travels as packets — small pieces of data (typically up to 1500 bytes each).

Step 5: How packets travel

A single page request might be split into many packets. Each packet:

• Has a source and destination IP address.
• Travels independently through the internet.
• May take a different physical path than its siblings.
• Gets reassembled at the destination.

The journey:

Your home network

Your device → router → modem → ISP cable.

ISP backbone

Your ISP routes the packet through its internal network using routers that decide the next hop based on routing tables.

Internet exchange points (IXPs)

When packets need to leave your ISP, they reach an Internet Exchange Point — physical locations where many networks connect. Major IXPs are in places like Frankfurt, Amsterdam, and Ashburn (Virginia).

Undersea cables

If the destination is on another continent, packets travel through fibre optic cables on the ocean floor. There are hundreds of these — between 95-99% of all transcontinental internet traffic flows through them. The big ones cost hundreds of millions of dollars to lay.

The destination network

Eventually the packet reaches the server's hosting provider, then the specific datacentre, then the specific server.

Step 6: The server responds

The web server processes the request and sends back HTML, CSS, JavaScript, and images. These too get split into packets and travel back through the same kinds of paths (often different physical routes — the internet doesn't guarantee symmetric paths).

Step 7: Your browser renders the page

Once all packets arrive, your browser:

• Reassembles them into the HTML document.
• Parses the HTML.
• Discovers it needs CSS, JS, images, fonts — sends additional requests.
• Builds the rendering tree.
• Paints the page on screen.

A typical modern web page makes 50-200+ requests just to render. Browsers parallelise heavily to speed this up.

The protocol stack

The internet is built on layers, each handling specific concerns:

Layer	Examples	Job
Application	HTTP, HTTPS, DNS	What the data means
Transport	TCP, UDP	Reliable delivery (or not)
Network	IP	Routing across networks
Link	Ethernet, Wi-Fi	Local network transmission
Physical	Fibre, copper, radio	Actual signal transmission

Each layer doesn't care how lower layers work; it just trusts them to do their job. This separation is why the internet can run on copper wires, fibre, satellites, and Wi-Fi simultaneously.

What is an IP address?

A unique identifier for a device on a network.

• IPv4 — 32-bit addresses (e.g. 76.76.21.21). About 4.3 billion possible. Mostly exhausted.
• IPv6 — 128-bit addresses (e.g. 2001:0db8:85a3::8a2e:0370:7334). Effectively infinite. Slow rollout but increasing adoption.

Most home users get a single public IPv4 from their ISP. Multiple devices share it via NAT (Network Address Translation).

How is the internet not centralised?

The internet has no headquarters. It's a federation of networks following common protocols. Key facts:

• No single owner — thousands of ISPs and networks interconnect voluntarily.
• No single point of failure — packets route around problems automatically.
• Standards bodies (IETF, ICANN) coordinate protocols and naming, but don't control infrastructure.

This decentralisation is why the internet survived the dot-com crash, regional outages, censorship attempts, and even physical cable cuts.

Common misconceptions

"The internet is the cloud"

Not the same. Internet = global infrastructure. Cloud = computing services you rent on the internet (AWS, Google Cloud, Azure).

"5G is internet"

5G is a wireless protocol for the last hop between your phone and the cell tower. The internet beyond that is still fibre.

"Internet is wireless"

The vast majority of internet traffic travels over physical fibre cables. Your Wi-Fi connection at home is a small final segment.

"Bigger pipe = faster internet"

Bandwidth and latency are different. A fat pipe with high latency feels slow for browsing. Most user-perceived speed depends on server distance and DNS speed, not raw bandwidth.

Internet Governance: Who Actually Controls It?

The internet has no CEO. Instead, governance is distributed across multiple organisations with specific responsibilities:

ICANN (Internet Corporation for Assigned Names and Numbers) manages domain name registration, IP address allocation, and root server operations. It's quasi-governmental — initially US-based, now with international representation — but disputes over its neutrality persist.

IETF (Internet Engineering Task Force) develops protocols (TCP/IP, DNS, TLS). It's open and merit-based — anyone can propose standards. The process is democratic but slow, which is why adding IPv6 support took decades.

Regional Internet Registries (RIRs) allocate IP addresses and manage regional internet policy. Five exist globally (ARIN for North America, RIPE NCC for Europe, LACNIC for Latin America, APNIC for Asia-Pacific, AFRINIC for Africa).

ISPs and backbone operators own the physical infrastructure. There's no central ISP; instead, thousands interconnect voluntarily at Internet Exchange Points.

This decentralisation is both a feature and a bug. Feature: no single entity can shut down the internet. Bug: nobody can unilaterally enforce standards or improve security globally.

Source: Mueller, M. (2010). Networks and States: The Global Politics of Internet Governance. MIT Press. — Historical analysis of internet governance institutions.

Network Resilience: Redundancy and Failure Recovery

How does the internet survive? Redundancy at every layer.

Path redundancy: If one route fails, packets instantly reroute. ISPs maintain multiple connections to multiple peers. A single cable cut in an ocean doesn't take down the continent because traffic reroutes around it.

DNS redundancy: The root servers are distributed across 13 globally redundant clusters. Even if several are offline, queries still resolve. Likewise, authoritative nameservers are replicated.

Server redundancy: High-traffic sites run in multiple datacentres. If one fails, traffic shifts automatically. This is what makes modern cloud infrastructure reliable.

Protocol-level robustness: TCP automatically retransmits lost packets. DNS automatically retries failed queries. HTTP includes timeouts and connection pooling. These automatic recovery mechanisms are why the internet feels reliable despite constant small failures.

The cost: redundancy adds expense. Every critical system runs in multiple locations, multiplying infrastructure costs. This is why internet infrastructure is expensive and why centralised systems (clouds) can be cheaper — they trade decentralisation for efficiency.

Source: Claffy, K.C., Clark, D., & Mohapatra, P. (2023). Rethinking Internet Resilience: Lessons from the COVID-19 Pandemic. IEEE Internet Computing, 27(2). — Empirical study of internet resilience during large-scale disruptions.

Scaling and Modern Internet Challenges

The internet today faces challenges the original designers didn't anticipate:

Congestion: A single popular video can consume gigabits per second. Without careful traffic engineering, one popular item can congest a link. CDNs (Content Delivery Networks) solve this by caching content near users, reducing backbone traffic.

DDoS attacks: Attackers flood servers with traffic, overwhelming capacity. Defenses include rate limiting, traffic filtering, and service provider mitigation (Cloudflare, Akamai offer DDoS protection).

IPv4 exhaustion: The 4.3 billion IPv4 addresses are nearly depleted. IPv6 solves this but requires all infrastructure to upgrade — a multi-decade transition underway since 1996.

Latency: The speed of light limits latency. A transatlantic fibre connection (5,500 km) has ~27 ms latency one-way — physics limits. High-frequency trading firms pay millions for minimal-latency cables to exchanges.

Energy consumption: Global internet infrastructure uses ~10% of electricity. Datacentres account for ~3%. As AI workloads scale, this will increase, pressuring providers to use renewable energy.

Source: Shehabi, A., Smith, S., Masanet, E., & Brown, R. (2021). Data Center Energy-Use Trends. Energy Efficiency, 14(7). — Quantification of internet energy consumption.

Security Across the Stack

The internet was designed for openness, not security. Cryptography was added later:

At the network layer (IP): No authentication — routers trust that IP addresses are legitimate. This is why IP spoofing and MITM (man-in-the-middle) attacks are possible. IPsec adds encryption but is complex and rarely deployed end-to-end.

At the transport layer (TCP): No encryption by default. TLS (Transport Layer Security) adds it, but must be explicitly used (HTTPS). The shift from HTTP to HTTPS took 20+ years; full coverage arrived only around 2020.

At the application layer: Each application must implement its own security. Email encryption (PGP) is rare, despite existing for 30 years. This is why email is fundamentally insecure.

DNS security: DNS responses are unsigned by default, enabling DNS spoofing. DNSSEC signs responses, but adoption remains partial (~40% globally).

The pattern: security is bolted on, not built-in. This works for user-initiated connections but fails for machine-to-machine communication and IoT, where security is often ignored for simplicity.

Source: Kaminsky, D. (2010). Black Ops 2010: It's the End of the Cache as We Know It. Security conference talk on DNS spoofing vulnerabilities. — Analysis of fundamental DNS security weaknesses.

Content Delivery Networks (CDNs): Moving Data Closer to You

The internet as originally designed was a peer-to-peer architecture: your browser connected directly to a web server. This works perfectly fine until a single server becomes popular. If millions of people try to watch the same video simultaneously, the origin server's bandwidth capacity collapses. The solution: distribute copies of the content geographically so that each user downloads from a server nearby.

Content Delivery Networks do exactly this. Companies like Cloudflare, Akamai, Fastly, and AWS CloudFront maintain thousands of edge servers distributed globally. When you request a video from Netflix or an article from a news site, the CDN's routing intelligence directs you to the geographically closest server that has a cached copy. If you're in Singapore, you download from a Singapore server; if you're in São Paulo, a São Paulo server. Latency drops from 100+ milliseconds to 10-20ms, and the origin server is protected from the thundering herd.

CDNs also perform smart caching: static content (images, CSS, JavaScript) is cached aggressively; dynamic content (personalized feeds, real-time data) is cached briefly or not at all. An originating server runs an HTTP header (Cache-Control: max-age=3600) telling CDNs how long to cache; CDNs respect this and re-fetch from origin only when the cache expires. For a high-traffic site, this means the origin server might receive 100 requests per second even though it's serving 100 million users — the CDN absorbs 99.9% of the load.

Source: Adhikari et al. (2012). "A Measurement Study of the Impact of Slow Links on Web Performance." IEEE/ACM Transactions on Networking, 20(2), pp. 373-386. — Study of latency reduction via CDN deployment.

Caching at Every Layer

The internet is built on a principle that sounds counterintuitive: repeat the same data as many times as possible. Caching happens at multiple layers to minimize repeated fetches:

Browser cache: Your web browser stores resources locally. The first time you visit a site, the browser downloads HTML, CSS, JS, images. The second time (within days or weeks), it loads them from disk — near-instant. Browser caches are so important that web developers spend enormous effort tuning cache headers to balance freshness (update when content changes) and efficiency (don't re-download).

ISP cache: Your ISP's DNS resolver caches DNS answers. If 1,000 users ask "what is youtube.com's IP?", the ISP resolver answers the first query by querying root servers, then caches the answer and responds to the next 999 from cache. This single optimization reduces load on root servers by roughly 99%.

CDN cache: As mentioned, CDNs cache content at edge locations to reduce latency and origin load.

Application-level cache: Web applications use Redis or Memcached to cache database queries. A user's profile might be queried 1,000 times per day; caching it in memory for 5 minutes means only 288 database hits instead of thousands. This is why Redis is a cornerstone of web-scale infrastructure.

Database indexes: At the lowest level, database engines cache frequently-accessed rows in memory (buffer pool), avoiding slow disk lookups.

All these caches operate on the same principle: locality of reference. Most traffic is concentrated on popular items (Zipfian distribution). Caching popular items near the requester multiplies throughput and reduces latency dramatically.

Source: Brewer et al. (2000). "Caching and Indexing for Content Delivery Networks"; modern data systems papers on cache invalidation strategies.

Network Neutrality and Internet Regulation

One of the most heated debates in internet governance is network neutrality — the principle that ISPs must treat all data equally, regardless of source, destination, or application. Net neutrality supporters argue that ISPs should not be allowed to prioritize their own services, block competitors, or charge content providers for "fast lanes." Opponents argue that allowing ISPs to manage traffic enables better network optimization and incentivizes infrastructure investment.

The regulatory landscape varies globally:

• European Union: Strong net neutrality rules via the 2016 Directive on electronic communications. ISPs cannot block or throttle legal content.
• United States: Pendulum swings based on administration. The FCC reclassified broadband as a "telecommunications service" (enforcing net neutrality, 2015), then reclassified as an "information service" (weakening rules, 2017), then back to telecommunications (2023). This regulatory uncertainty makes long-term planning difficult.
• China: No net neutrality concept; the government manages internet access and content through the Great Firewall.
• India: Strong net neutrality protections since 2015, though enforcement remains inconsistent.

Why it matters: If ISPs could create "fast lanes," startup companies without deep pockets would be disadvantaged. The internet's openness — the feature that enabled YouTube, Netflix, and Airbnb to succeed without ISP permission — depends on net neutrality. Conversely, ISPs argue that without the ability to manage traffic (prioritizing video delivery, for instance), networks degrade for everyone. The optimal policy remains contested.

Source: Labovitz et al. (2010). Does packet sampling bias network delay measurements? IEEE/ACM TON; also FCC proceedings and BEREC (Body of European Regulators) guidance documents on net neutrality (2022-2024).

The Future of Internet Infrastructure

The internet we have now faces pressure from multiple directions. Solutions are being deployed and researched:

IPv6 adoption: IPv4 is nearly exhausted. IPv6 provides ~340 undecillion addresses. Adoption has been slow (roughly 40% global adoption as of 2024), partly due to lack of necessity (NAT and IPv4 scarcity somehow work), partly due to stubborn legacy infrastructure. As IoT and edge devices proliferate, IPv6 becomes essential, and adoption is accelerating.

Quantum computing threat: Current encryption (TLS, which uses RSA or elliptic curves) is vulnerable to future quantum computers. NIST is standardizing "post-quantum cryptography" algorithms designed to resist quantum attacks. Migration will take a decade or more; early birds are already deploying hybrid systems (both classical and post-quantum crypto simultaneously).

Latency improvements: The speed of light limits latency on long distances. Fiber-optic cables are being laid continuously to reduce latency for high-frequency trading and interactive applications. Some firms invest in specialized ultra-low-latency networks that bypass the public internet for financial transactions. For most users, current latency (30-50ms within country, 100-200ms transcontinental) is acceptable; for specialized applications (VR, gaming, real-time collaboration), further reduction drives infrastructure development.

Decentralized internet: Some activists advocate for mesh networks, Bitcoin.">blockchain-based naming systems, or federated protocols to reduce dependence on centralized ISPs and cloud providers. Practicality remains limited; decentralized systems are slower, harder to secure, and require user sophistication. Progress exists (IPFS, Mastodon federation) but mainstream adoption is slow.

Satellite internet: SpaceX's Starlink and Amazon's Project Kuiper offer global coverage, bringing internet to remote areas where fiber is prohibitively expensive. Latency is high (~20-50ms, compared to fiber's 5-10ms) and bandwidth is limited (100-500 Mbps) compared to fiber (gigabits/s). But for regions with no alternatives, satellite is transformative. Scaling to billions of users remains an open question.

Source: NIST Post-Quantum Cryptography standardization documents (2024); ITU report on broadband adoption by region (2024); Starlink white papers on latency and capacity.

The Internet's Carbon Footprint

Global internet infrastructure accounts for roughly 2-3% of CO₂ emissions — comparable to aviation. Datacentres alone consume about 1% of global electricity. As AI workloads scale (training large language models can consume gigawatt-hours), energy demand is increasing.

Mitigation strategies underway:

• Renewable energy: Major cloud providers (Google, Microsoft, Amazon) commit to 100% renewable energy for datacenters by 2030-2035. Progress is real but incomplete; existing datacenters run on mixed grids.
• Efficiency improvements: Modern cooling techniques (liquid cooling instead of air), neuromorphic chips, specialized accelerators (TPUs, GPUs designed for specific workloads) reduce energy per compute unit.
• Carbon-aware routing: Some ISPs and cloud providers adjust traffic routing to use grids with lower carbon intensity at specific times. This is nascent but growing.
• End-user device efficiency: Shifting computation to edge devices (your phone running models locally instead of sending queries to the cloud) can reduce transmission energy, though device compute still consumes power.

The internet as currently built is not sustainable at massive scales. Continued growth requires decoupling from fossil fuels — a challenge that requires both technology and energy policy alignment.

Source: Shehabi et al. (2021). Data Center Energy-Use Trends; Kaur & Sharma (2019). "Energy Efficiency in Data Centers: A Review."

The takeaway

The internet is a remarkable example of how decentralised systems can be more resilient than centralised ones. It survives because nobody controls all of it. Billions of devices coordinate using protocols designed 50 years ago, yet the system evolves and scales. When you click a link, an invisible ballet of DNS queries, TCP handshakes, packet routing, and TLS encryption unfolds in milliseconds. None of the routers between you and the destination server know what the page means; they just move bits as instructed. The architecture is so robust that most internet users never notice the thousands of failures, reroutes, and retransmissions happening constantly in the background. Yet the internet also faces profound challenges: exhausting IP addresses, evolving threats from quantum computing, energy consumption that rivals industries, and regulatory uncertainty. Solutions exist but require sustained investment, coordination, and global agreement on principles — a challenge as complex as the infrastructure itself.

Frequently Asked Questions

Q: Who owns the internet?
A: No single entity owns it. Different parts are owned by ISPs, governments, companies, and universities. Protocols and naming systems are governed by international bodies (ICANN, IETF) with no owner, only coordinating stewards.

Q: Where is the internet physically located?
A: Everywhere and nowhere. Datacentres worldwide, undersea fibre-optic cables, cell towers, home routers, and millions of servers. The "cloud" is not virtual — it's physical infrastructure distributed globally.

Q: Can the internet be turned off?
A: Locally or regionally, yes — through DNS blocking, border filtering, or infrastructure destruction. Globally, no — the network is too distributed. Even massive outages (cable cuts, datacentre failures) affect regions, not the global internet.

Q: How do undersea cables function?
A: Fibre-optic cables on the ocean floor with periodic repeaters that boost signals. Typical lifespan: 25 years. Damage (anchor drags, trawlers, rarely sharks) causes occasional regional outages. Specialised cable ships repair breaks within hours to days.

Q: Why does my Wi-Fi feel slow even though my internet speed is fast?
A: Local issues (interference, distance from router, channel congestion, device limits) often dominate. Speed test tools show whether bottlenecks are local (Wi-Fi) or upstream (ISP). Wi-Fi operates on shared radio spectrum; many devices can congest it.

Q: How long did it take to build the internet?
A: ARPANET (the prototype) started 1969. Commercial internet began ~1990. The global infrastructure of today took 50+ years to build, driven by academic research, commercial competition, and regulatory environment changes.

---

Read next:

• How WiFi Actually Works — The Physics of Wireless Signals
• How Cloud Computing Actually Works
• How Undersea Internet Cables Work — The Invisible Infrastructure