Tiered caching is a content delivery architecture in which caches are organized in two or more hierarchical layers, so that an edge cache that misses a request asks an upper-tier parent cache before contacting the origin server. In Content delivery across the Internet, it raises the overall cache hit ratio and sharply reduces origin traffic, because many edge locations share a smaller set of well-populated upper caches.
The idea dates back to hierarchical proxy caching, but modern CDNs apply it at global scale. The trade-off is an extra network hop on misses in exchange for fewer expensive trips to the origin.
How tiered caching routes a request
Tiered caching routes each request through a fixed lookup order, stopping at the first layer that holds a fresh copy. A client first reaches a lower-tier edge cache at a nearby point of presence. On a miss, that edge does not contact the origin; it forwards the request to an assigned upper-tier cache, often located in a large data center with strong connectivity to the origin.
If the upper tier holds the object, it returns it to the edge, which stores a copy for later clients. Only when the upper tier also misses does a request travel to the origin server. Because dozens of edges funnel into one parent, the parent sees the combined demand of a region and keeps popular objects warm far more often than any single edge could.
The following table lists the typical layers and their roles.
| Layer | Typical location | Role | Cache size |
|---|---|---|---|
| Lower tier (edge) | Point of presence near users | Serves clients with low latency | Smaller, high churn |
| Middle tier | Regional hub | Aggregates misses from many edges | Medium |
| Upper tier (parent) | Large data center near origin | Last cache before the origin | Largest, long retention |
| Origin server | Customer infrastructure or storage | Authoritative source of content | Not a cache |
Tiered caching topologies
Tiered caching topologies differ mainly in how edges are mapped to parents. In a regional topology, each edge uses the nearest upper-tier location, which keeps miss latency low but spreads the same object across several parents. In a single-parent or smart topology, the CDN chooses one parent for each origin based on measured latency, so the content is concentrated in one place and the hit ratio rises further.
Some providers also hash content across several parents, so that each URL has a deterministic home cache. This consistent hashing avoids duplicate copies and scales storage, but a failure of one parent shifts its share of objects elsewhere and causes a temporary miss storm.
The comparison below shows how the main topologies balance hit ratio, latency and resilience.
| Topology | Edge-to-parent mapping | Hit ratio effect | Main drawback |
|---|---|---|---|
| Flat (no tiers) | Every edge contacts the origin | Lowest | Heavy origin load |
| Regional tiers | Nearest regional parent | Moderate to high | Duplicate copies per region |
| Single parent | One parent chosen per origin | Highest | Long miss path for distant edges |
| Hashed parents | URL hash selects parent | High | Rebalancing after failures |
Tiered caching compared with an origin shield
Tiered caching and an origin shield overlap heavily, and the terms are often used interchangeably. Strictly, an origin shield is a single designated cache layer placed directly in front of the origin, usually configured per customer. Tiered caching is the broader architecture of multiple cache layers, of which a shield is one specific implementation.
Both differ from a multi-CDN setup, which spreads traffic across independent providers and usually lowers the hit ratio because each provider warms its own caches. Request collapsing, where an upper tier merges simultaneous misses for the same object into one origin fetch, complements tiering and is often enabled together with it.
| Concept | Scope | Effect on origin load | Typical configuration |
|---|---|---|---|
| Tiered caching | Hierarchy of cache layers | Strong reduction | CDN-wide or per zone |
| Origin shield | One layer before the origin | Strong reduction | Chosen shield location |
| Edge cache only | Single layer at the edge | Limited reduction | Default CDN behavior |
| Multi-CDN | Several independent providers | Often increases | DNS or traffic steering |
When tiered caching pays off
Tiered caching delivers the most value for cacheable content with a long tail of moderately popular objects. Large media libraries, software downloads, image catalogs and video segments benefit because each edge sees too few requests per object to stay warm, while the parent sees enough. Gains of tens of percentage points in origin offload are commonly reported, although the exact figure depends on traffic distribution and TTL settings.
Correct Cache-Control headers remain a precondition. Tiering cannot help with responses marked private or no-store, and short TTLs force frequent revalidation through every layer. Directives such as stale-while-revalidate let tiers serve slightly stale content while refreshing it in the background.
| Situation | Recommendation | Reason |
|---|---|---|
| Global audience, single origin | Enable tiered caching | Many edges share one warm parent |
| Highly dynamic, uncacheable API | Little benefit | Responses bypass the cache |
| Origin with egress fees | Enable with request collapsing | Fewer origin fetches lower cost |
| Users in one city only | Optional | Few edges, little aggregation |
Limitations and operational pitfalls
The main cost of tiered caching is added latency on a full miss, since the request crosses an extra hop before reaching the origin. For edges far from their parent, this can add tens of milliseconds to the first request for an object.
- A cache purge must invalidate every tier; otherwise an upper tier may refill edges with outdated content.
- Inconsistent cache keys between tiers, such as differing query string handling, cause silent misses.
- A failed parent can trigger a miss storm against the origin if no fallback parent exists.
- Edge-level analytics may overstate origin hits, because upper-tier hits look like misses from the edge perspective.
Understanding it simply
Imagine a chain of small corner shops supplied by one factory. Without a warehouse, every shop that runs out calls the factory directly. With a regional warehouse in between, shops restock from the warehouse, and the factory only ships when the warehouse itself runs out.
Tiered caching works the same way for web content. Nearby servers ask a larger shared server first, so the original website receives far fewer requests and stays fast and affordable to run.
Applied in practice
An online learning platform stores about 40 TB of course videos on a single cloud storage origin in Europe, with students on several continents. Before tiering, each of roughly 200 edge locations fetched video segments independently, and the origin served about 35 percent of all requests, generating high egress charges.
The team enables tiered caching with one upper tier near the storage region and turns on request collapsing. Edges in Asia and South America now fill from the parent instead of the origin. Within a week, the origin share drops to under 8 percent of requests, egress costs fall accordingly, and the first-play delay rises only slightly for rarely viewed lessons.