Understanding CDN caching#

CDN stands for Content Delivery Network. The basic principle of CDN is to widely use various caching servers, distributing these caching servers to regions or networks where user access is relatively concentrated. When users access a website, global load balancing technology is used to direct their access to the nearest functioning caching server. The caching server directly responds to user requests. The basic idea of CDN is to avoid potential bottlenecks and issues that may affect data transmission speed and stability on the internet, making content delivery faster and more stable. By placing node servers in various locations on the network, CDN systems can dynamically redirect user requests to the nearest service node based on network traffic, node connections, load conditions, distance to users, response time, and other comprehensive information. The goal is to allow users to obtain the required content nearby, solve congestion issues on the internet, and improve the response speed of accessing websites.

Components#

• Functionally, a CDN system consists of a distribution service system, a load balancing system, and an operation management system. The distribution service system is mainly responsible for resource response, caching, and synchronization. The load balancing system is mainly responsible for balancing the load of multiple content caching devices and caching load balancing, as well as access control and routing of content. The operation management system is responsible for operational requirement management and network system management.
• In terms of node distribution, the CDN system is mainly divided into the edge layer and the central layer. The edge layer is distributed at the edge of the CDN network, providing users with nearby access services. The central layer is responsible for resource synchronization and operation management functions. The central layer stores relevant configuration information for accelerated domains, such as the origin domain name, and also caches various resources under the accelerated domain. When the edge layer node fails to hit the cache, it needs to send a request to the central layer node. If the central layer node fails to hit the cache, it needs to look up the corresponding origin domain name and send a request to that origin domain name, and then return and cache the requested resource layer by layer.

Functions#

• Save backbone network bandwidth and reduce bandwidth demand.
• Reduce the impact of communication storms and improve the stability of network access.
• Provide server-side acceleration to solve server overload issues caused by high user traffic.
• Overcome the problem of uneven distribution of website users and reduce the construction and maintenance costs of websites.
• Provide resource access caching to reduce response latency for accessing the same object and reduce backbone network bandwidth usage.

Key Technologies#

• Cache algorithms determine hit rates, source server load, and POP node storage capacity.
• Distribution capability depends on IDC capability and IDC strategically distributed.
• Load balancing determines optimal routing, response time, availability, and service quality.
• DNS-based load balancing achieves optimal node service through CNAME.
• Cache points include client browser cache and local DNS server cache.
• Cache content includes DNS address cache, client request content cache, and dynamic content cache.
• Supported protocols include static and dynamic acceleration, image acceleration, HTTPS certificate acceleration, download acceleration, and more.

Configuration#

When using CDN services provided by a CDN service provider, some configurations need to be made:

• Resolve a subdomain, which can be initially resolved to any address, such as cdn.example.com.
• Add this domain to the service provider and set the origin domain name, such as www.example.com.
• At this point, the service provider will generally assign a CNAME address, such as cdn.example.com.service.com.
• Add the domain from the first step as a CNAME record to the assigned CNAME address.
• Alternatively, if the service provider provides a CNAME address in the first step, simply resolve it.

Access Process#

The access process of a simple CDN, using a pull method to fetch the cache, is as follows:

• When accessing a resource, load the resource file from the aforementioned subdomain, and perform DNS resolution for that domain.
• Return the CNAME address, and then resolve the CNAME address.
• Obtain the IP address corresponding to the CNAME domain name, pointing to the CDN edge layer node.
• If the CDN edge layer node fails to hit the resource cache, it requests from the central layer node.
• If the central layer node fails to hit the resource cache, it retrieves the resource from the origin domain name server.
• After successfully obtaining the resource, it is returned layer by layer and cached.
• During this process of searching for resources, the domain name may change, but the resource path remains unchanged.
• Subsequent accesses can directly retrieve the cache from the edge node without the need to retrieve from the origin, speeding up resource access.

Cache Control#

There are two major challenges in computing: when the cache expires and how to name it. In the context of CDN, cache expiration is a tricky problem. If the resource files on the origin server change and the user retrieves the resources from the cache node, it will result in inconsistent resource files. One way to solve this problem is to refresh all CDN caches by actively pushing them, but this method is costly. A simpler solution is to invalidate the cache after a fixed period of time. In addition to controlling the cache on the nodes, the user's local cache also needs to be controlled. The HTTP protocol provides the following cache control methods:

Strong Cache#

Strong cache controls the validity period of the cache stored locally through Expires and Cache-Control.

Expires#

Expires is a header introduced by HTTP 1.0 to indicate the expiration time of a resource. It represents an absolute time returned by the server. Expires is limited by the local time, so modifying the local time may cause the cache to expire. For a resource request, if it is within the Expires period, the browser will directly read the cache and will not request the server.

Cache-Control#

Cache-Control appeared in HTTP 1.1 and has a higher priority than Expires. It represents a relative time and is supported by both request and response headers. Different values provided by Cache-Control define different caching strategies.

• Cache-Control: no-store: The cache cannot store any content related to client requests and server responses. Each request initiated by the client will download the complete response content.
• Cache-Control: no-cache: The cache stores the server's response content, but this cache cannot be provided to the browser until the freshness is revalidated with the server. In simple terms, the browser caches the server's response resources, but for each request, the cache needs to evaluate the validity of the cached response with the server. Based on whether the response is 304 or 200, it determines whether to use the local cached resource or the server's response resource.
• Cache-Control: public || private: public indicates that the response can be cached by any intermediary, such as a proxy or CDN. The default response is private, which means that the response is private and intermediaries cannot cache it. The response can only be applied to the browser's private cache.
• Cache-Control: max-age=31536000: The response has the maximum expiration time, and the directive is max-age=<seconds>, which indicates the maximum time the resource can be cached and kept fresh, in seconds from the time the request is initiated.
• Cache-Control: must-revalidate: When the must-revalidate directive is used, it means that when the cache considers using a stale resource, it must first validate its status. Expired caches will not be used. In normal cases, it is not necessary to use this directive because in the case of strong cache expiration, negotiated caching will be performed. However, the HTTP specification allows clients to use expired caches in certain special cases, such as when the validation request fails or when there are some special directives configured, such as stale-while-revalidate and stale-if-error. The must-revalidate directive ensures that the cache must be revalidated in any case after expiration.

Negotiated Caching#

When the browser's request for a resource does not hit the strong cache, it sends a request to the server to verify if the negotiated cache is hit. If the negotiated cache is hit, the response status returned is 304 (Not Modified), and the request does not carry entity data. If it is not hit, the response is 200 and carries the resource entity data. Negotiated caching is managed using the Last-Modified, If-Modified-Since and ETag, If-None-Match pairs of headers.

Last-Modified If-Modified-Since#

Last-Modified and If-Modified-Since were introduced in HTTP 1.0. Last-Modified indicates the last modification date of the local file. The browser adds If-Modified-Since to the request header, which is the value of the previous response's Last-Modified, to ask the server if the resource has been updated since that date. If there are updates, the server will send the new resource back. However, if the cached file is opened locally, the Last-Modified will be modified. Therefore, ETag was introduced in HTTP 1.1.

ETag If-None-Match#

ETag is like a fingerprint. Any changes to the resource will cause the ETag to change, regardless of the last modification time. ETag ensures that each resource is unique. The request header field If-None-Match sends the previously returned ETag to the server to inquire if the resource's ETag has been updated. If there are changes, the server will send the new resource back. ETag has a higher priority than Last-Modified. The use of ETag is mainly considered in the following situations:

• Some files may change periodically, but their content remains the same, such as only modifying the modification time. In this case, we do not want the client to think that the file has been modified and re-GET it.
• Some files are modified very frequently, such as multiple times within a second, for example, N modifications within 1s. The granularity that If-Modified-Since can detect is in seconds, so it cannot detect such rapid modifications.
• Some servers cannot accurately obtain the last modification time of a file.

Daily Question#

References#