Why is CDN designed this way?

Over the past few decades, computer networks have connected almost all the computers in the world. We only need to deploy static resources and dynamic codes to the server, and then start the service to listen to a certain port, so that computers around the world can access the website.

But there is a problem with this. Resources are ultimately transmitted through physical layer network lines and devices. It takes some time to pass through each line and each network device, so the farther the client and server are apart, the slower the website will open.

It's like you bought something from Hainan. If you are in Guangzhou, you may receive it very quickly because the transmission distance is short, but if you are in Beijing, you may have to wait a few more days because there are more lines and nodes in between.

But this is definitely not feasible, the user experience will be very poor. How to solve this problem?

The farther away the website is, the slower it will open. It is easy to imagine that if it is deployed in many places, when users access the network, wouldn’t it be enough to access the nearest one?

It's just like express delivery has some transit warehouses that can store some goods. If you are in Beijing and want to buy something from Hainan, and it happens to be in the warehouse in Beijing, then you can receive it very quickly.

The idea is fine, but how to implement it?

Users access websites through domain names, so can this function be achieved through DNS servers?

I wrote an article about the principles of DNS before, so here is a brief review:

When a client accesses a domain name, it will first search the local hosts file, and if the IP address can be found, it will access it directly.

Otherwise, it will send a request to the local DNS server, which is provided by operators such as China Unicom and China Mobile in every city. It will send a request to the domain name server to resolve the domain name, and then return the result to the client.

Domain names are resolved in layers, with three layers: root domain name server, top-level domain name server, and authoritative domain name server. For example, image.baidu.com will first send a request to the root domain name server to query the IP address of the top-level domain name server of com, and then query the IP address of the authoritative domain name server of image.baidu.com from the top-level domain name server of com. After querying the authoritative domain name server, domain names at any level will be resolved here (so it is called the authoritative domain name server).

When you see this authoritative domain name server, I wonder if you have thought about how to implement the CDN network.

Can we do load balancing based on the client IP at the authoritative domain name server level? For example, a DNS request from Beijing will return the IP of the server in the Beijing data center, and a DNS request from Shanghai will return the IP of the server in the Shanghai data center.

This is indeed possible to achieve local content distribution, and such a load balancing network is called CDN (Conent Delivery Network)

However, to implement such a CDN network, it is necessary to build multiple computer rooms across the country, which is too costly. Therefore, only large companies like Baidu, Alibaba, and Tencent will build their own CDNs. Generally, we will buy third-party CDN services.

These companies have built CDN networks, which they actually cannot fully use themselves, and they also provide CDN acceleration services to the outside world.

Third-party CDN services naturally also need to provide a DNS server, which is the one that returns the IP addresses of servers in different cities based on the IP addresses.

For example, this is the schematic diagram of Baidu Cloud CDN:

After the user sends a request to the local DNS server, it will go through the DNS resolution of the root domain name and top-level domain name, and will eventually be forwarded to the authoritative DNS server. At this time, the authoritative DNS server only needs to forward it to Baidu's DNS server, so that the CDN service can be connected.

Baidu's DNS server implements load balancing, and will return the IP addresses of servers in different cities based on the city where the requesting IP address is located, thus realizing the network acceleration function of nearby distribution.

So how is the forwarding from the authoritative DNS to Baidu's DNS achieved?

There are many types of DNS records, such as:

A stands for address, which records the IP address corresponding to the domain name.

CNAME means that the domain name also has an alias, and the IP address can be checked against that domain name.

MX represents the domain name or IP corresponding to the suffix of the name

When you see this CNAME type, you should know how to implement forwarding.

Just configure a CNAME record on your DNS server and point it to the domain name of the CDN server.

For example, when you use a certain cloud CDN, the first step is to configure your own DNS server's CNAME to point to it:

In this way, when you visit a domain name, the authoritative server that resolves the domain name will return the domain name of the CDN service's DNS server, and then send a request to resolve the domain name to the CDN's DNS server. At this time, it can return a server in the nearest city to you based on the city where the IP is located.

Of course, you can also do another layer of CNAME forwarding, for example, the CDN DNS server transfers the domain name resolution to the city's DNS server, and then the city's DNS server returns the IP address of a server that is closer and has a smaller load to the client based on the load conditions of different machines.

This allows the client to download static resources from the nearest server, allowing the website to open faster.

If the accessed resource does not exist, a request will be sent to the origin server to obtain the corresponding resource and cache it. The origin server does not need to be accessed for subsequent access.

How is this cache time specified?

The CDN service will have a console that can set the cache time of different resources. Of course, there are also expressions and Cache-Control in the request header to control the cache time, and the priority of these two can also be set.

Such CDN services are separate for domestic and foreign users. If the website's target users are only domestic, then you only need to purchase domestic CDN services. If foreign users have slower access, then it's fine. If the website has a lot of foreign users, then you also need to purchase foreign CDN services.

For example, the domestic CDN node distribution of a certain cloud is as follows:

Overseas, there is another CDN network:

Summarize

In order to speed up the opening of the website, we will use CDN service, which is not a network protocol, but a network with acceleration function based on DNS protocol.

Its principle is that the authoritative DNS server of the domain name forwards the request to CND's load-balancing DNS server, and then returns the DNS servers in different cities based on the IP, and then selects the IP of a nearby server to return based on the load.

This way the client can get resources from the nearest server with the least load.

The CDN cache settings can be based on the header, or you can set some rules in the console, and the priority of the two can also be set.

The construction cost of CDN network is still very high. Only large companies will build it themselves. Generally, we will buy third-party services, and domestic and overseas CDNs are separate. If you mainly serve domestic users, you only need to buy domestic CDN services.

With this nationwide and global CDN node network, we can open the website faster.