AI 原生开放教材

外观

Easy-Vibe（Datawhale：从零做出真实产品）课时教程·中英混排原文 ↗原件 ↓仓库 ↗

2. CDN: Your "Global Courier Network" ​

2.1 Motivation for needing a CDN ​

Imagine you run an online store with servers in Shenzhen. Now a user in Beijing wants to access your images:

  • Without CDN: The request travels from Beijing → Hebei → Henan → Hubei → Hunan → Guangdong → Shenzhen, spanning over 2,000 kilometers—a round trip of over 4,000 kilometers. Network transmission alone takes tens of milliseconds, and it's even worse during network congestion.

  • With CDN: The request goes directly from Beijing to a Beijing CDN node (possibly in a Beijing Unicom data center). The distance shrinks from 2,000 km to 20 km, and latency drops from 50ms to 5ms.

This is the core value of CDN: bringing content closer to the user.

2.2 CDN Core Architecture ​

Edge Nodes: The "Courier Stations" Closest to Users ​

Edge nodes are the tier closest to users in the CDN network, typically deployed at:

  • ISP data centers (China Unicom/China Telecom/China Mobile)
  • Major city internet exchange points
  • Key transportation hubs

CDN Node Distribution in China:

  • Tier-1 cities: Beijing, Shanghai, Guangzhou, Shenzhen
  • Tier-2 cities: Hangzhou, Nanjing, Chengdu, Wuhan, Xi'an
  • Overseas: Hong Kong, Singapore, Tokyo, Silicon Valley, Frankfurt

Origin Server: The "Main Warehouse" of Content ​

The origin is where the CDN fetches content from when cache misses occur. It can be:

  • Object storage (OSS/COS/S3)
  • Self-managed servers (ECS/physical machines)
  • Load balancers (SLB/CLB)

Key Configuration:

  • Origin HOST: The domain/IP used by CDN nodes when accessing the origin
  • Origin Protocol: HTTP or HTTPS
  • Origin Port: 80, 443, or a custom port

Mid-Tier Nodes: "Regional Distribution Centers" ​

Between edge nodes and the origin, CDNs typically have one or more layers of mid-tier nodes:

  • Aggregation nodes: Consolidate origin requests from multiple edge nodes to reduce origin pressure
  • Regional centers: Responsible for content distribution and scheduling within a large region

Benefits of this layered architecture:

  1. Reduces origin pressure: 1,000 edge node requests may only require 10 requests to the origin
  2. Improves hit ratio: Popular content is intercepted at the mid-tier, no need to go back to the origin
  3. Fault isolation: If one link fails, traffic can automatically switch to another path

2.3 Complete CDN Acceleration Flow ​

Let's trace a real user request:

Step 1: DNS Resolution (Intelligent Scheduling)

User enters: cdn.example.com/image.jpg
↓
DNS server returns: Beijing Unicom CDN node IP (1.2.3.4)

The key here is intelligent DNS: based on the user's ISP, geographic location, and node load, it returns the optimal CDN node IP.

Step 2: Edge Node Lookup (Cache Hit?)

Request arrives at Beijing Unicom CDN node (1.2.3.4)
↓
Node checks local cache:
├─ Hit? Return content directly ✓
└─ Miss? Continue to next step

Step 3: Origin Fetch (Layer by Layer Upward)

Edge node miss
↓
Request to parent node (e.g., North China regional center)
├─ Parent node hit? Return content
└─ Parent node miss? Continue upward
    ↓
    Request to origin
    ↓
    Origin returns content

Step 4: Cache and Return (Faster Next Time)

Content returns along the chain
↓
Each layer caches a copy
↓
Finally reaches the user

This way, the next time a user requests the same file, it can be served directly from the edge node, achieving "instant loading."