Content Delivery Network

When you open a webpage, images, text, and videos have to come to you from somewhere. Usually they sit on a single computer somewhere on the internet, which sends them out on request. If that computer is in California and you’re sitting in Hamburg, the signal has to travel halfway around the globe. A Content Delivery Network solves this by distributing copies of the content across many computers worldwide. Your request then automatically lands on the copy that’s closest to you. In short: the internet doesn’t get faster, but the path gets shorter.

Why half a second decides on money

Data travels very fast, but not infinitely fast. A signal from Germany to California and back takes around 150 milliseconds, purely because of the distance. That sounds like little, but a single webpage often loads dozens of files one after another. These delays add up to noticeable waiting times.

For companies, this is directly measurable in euros. Online shops have observed for years that purchase numbers drop when pages get slower. With streaming services the effect is even harsher: if a video stutters, viewers drop off. That’s why practically every major provider uses a CDN, often even several at once.

Then there’s resilience. If a popular video suddenly attracts millions of viewers, a single server would collapse. If the load is spread across hundreds of locations, the system holds up. This also protects against attacks in which someone deliberately floods a page with requests.

Copies at the network edge and the problem with outdated content

A CDN operates servers in many cities, often directly at the major internet hubs. These locations are called edge servers, meaning servers at the edge of the network, close to the users. They store copies of frequently requested files in a temporary store, the cache. If someone requests an image that’s already in the cache, it’s delivered from there immediately.

If the file isn’t available yet, the edge server fetches it once from the actual server. This origin server is called the origin. After that, the edge server keeps the copy and serves all further requests itself. A well-configured CDN thus answers over 90 percent of all requests without ever bothering the origin.

The assignment to the nearest location usually happens via the Domain Name System, the internet’s address directory. Depending on where the request comes from, it returns a different server address. The biggest problem here is outdated copies: if a price changes in the shop, old versions may still sit in hundreds of caches. Operators control this through expiration times for each file and through targeted deletion commands.

Netflix, Cloudflare, and the role in AI services

You use CDNs every day without noticing. Every YouTube video, every Instagram image, and every software update on your phone runs through such networks. Netflix has even built its own devices for this and places them directly in the data centers of internet providers. The series you start watching in the evening is then often already sitting in the same city.

In business news, it’s mainly the operators that show up: Cloudflare, Akamai, Fastly, as well as the CDN divisions of Amazon, Google, and Microsoft. You notice their importance when something goes wrong. During outages at major CDN providers, thousands of well-known websites have already gone offline simultaneously on several occasions. Such outages show how heavily the network depends on a few companies.

CDNs are also playing a growing role in AI services. The actual computational work of a language model happens in large data centers, but the requests run over the same distributed infrastructure. Some providers now even run smaller models directly on the edge servers. Incidentally, a CDN is not hosting: it only stores copies, the original remains elsewhere.