What is Pull CDN?
A pull CDN retrieves content from the origin server on demand when a user requests it, storing the response for subsequent requests.
A pull CDN is a content delivery network architecture where edge servers fetch content from the origin server only when a user request arrives and the content is not already cached. The origin server remains the authoritative source, and the CDN acts as a transparent intermediary that populates its cache dynamically based on actual traffic patterns. This contrasts with a push CDN, where content is proactively uploaded to edge nodes before any user requests.
When a client requests a resource from a pull CDN, the edge node checks its local cache. On a cache miss (the resource is not present or has expired), the edge node sends a request to the origin server, retrieves the full response, stores a copy according to its caching policy (usually defined by Cache-Control headers), and then forwards the response to the client. Subsequent requests for the same resource from any client served by that edge node can be served from the cache until the TTL expires. Many pull CDNs also support purging or prewarming to manage cache state.
Pull CDNs are widely used for static web assets like images, CSS, JavaScript files, and even dynamically generated content that can benefit from caching (e.g., API responses with appropriate cache headers). They simplify origin management because content does not need to be manually or automatically distributed to every edge node. However, the first user for any given resource may experience higher latency (a cache-miss penalty). Pull CDNs dominate the market for general-purpose web acceleration, with major providers including Cloudflare, Akamai, Fastly, and AWS CloudFront operating primarily in pull mode for customer origin content.
Key facts
- Content is fetched from the origin only on cache miss at the edge node.
- Cache-miss penalty adds one origin round-trip time to the first request.
- No pre-seeding of content is required; origin traffic grows with cache misses.
- Cache expiration and invalidation are controlled by origin response headers.
- Pull CDNs are the default model for most cloud and reverse-proxy CDNs.
How it works in practice
Related terms
References
More in CDN & Performance
Apdex Score
The Apdex Score is a standardised metric that measures user satisfaction with application performance by comparing response times against predefined target and tolerable thresholds.
Bandwidth Throttling
Bandwidth throttling is the intentional reduction of data transfer speed by a network operator or service provider to manage congestion, enforce usage policies, or control costs.
Brotli Compression
Brotli is a lossless compression algorithm developed by Google, offering higher text compression ratios than gzip, used by CDNs to reduce page load times.
Cache Hit
A cache hit occurs when a requested resource is found in a CDN edge cache and served directly to the client, bypassing the origin server entirely.
Cache Invalidation
Cache invalidation is the explicit removal of stored web objects from a cache so that new requests must revalidate or refetch them from the origin server.
Cache Miss
A cache miss occurs when a requested resource is not found in a CDN or proxy cache, forcing the request to be forwarded to the origin server and then storing the response for future requests.
CDN
A CDN (Content Delivery Network) is a geographically distributed network of proxy servers and data centers that deliver web content to users from the nearest edge location, reducing latency and offloading origin servers.
Core Web Vitals
Core Web Vitals are a set of three real-world user experience metrics (LCP, INP, CLS) defined by Google to quantify loading, interactivity, and visual stability on web pages.
Cumulative Layout Shift
Cumulative Layout Shift (CLS) is a Core Web Vital metric that measures the sum of all unexpected layout shift scores during a page's lifespan, quantifying visual stability.
Edge Computing
Edge computing is a distributed computing model that processes data and runs application logic at Points of Presence (PoPs) close to end users, minimizing round-trip latency and bandwidth usage compared to centralized cloud regions.