Origin shielding and tiered caching can reduce repeated requests to your origin, but the extra layer is not automatically worthwhile. This guide shows how to compare direct CDN delivery, regional cache tiers, and a dedicated shield region using repeatable inputs such as cache-hit rates, origin bandwidth, request volume, latency, and provider pricing. Use the method during capacity planning, CDN migrations, and post-change cost reviews.
Overview
A conventional CDN serves a request from an edge location when the object is cached there. On a miss, that edge location retrieves the object from the origin. With origin shielding or tiered caching, the edge usually checks an intermediate cache first. Only a miss at that intermediate layer travels to the origin.
Terminology varies by provider. “Origin shield,” “shielding,” “tiered cache,” “parent cache,” and “regional cache” can describe related designs, but they are not always identical. Confirm the exact request path, supported shield regions, cache behavior, and billing model in your provider’s documentation before making a cost estimate.
The central trade-off is simple:
- Direct origin delivery: fewer components and potentially the shortest path for an origin that is close to most users.
- Regional tiering: multiple edge locations share a nearby parent cache, reducing duplicate origin fetches.
- Origin shielding: a selected shield region becomes the main gateway to the origin, concentrating misses and often making origin traffic more predictable.
Shielding is most useful when many edge locations request the same cacheable objects, the origin is costly or capacity-constrained, or traffic is geographically distributed. It may be less attractive when content is highly personalized, objects have very short lifetimes, or the shield adds a significant network path for frequent misses.
Before changing the design, review your existing cache behavior with the website cache audit checklist. A tiered cache cannot compensate for incorrect cache-control headers, unstable cache keys, frequent purges, or an application that marks reusable content as private.
How to estimate
Build a comparison using the same traffic period for each architecture. A full month is useful for billing analysis, while a representative busy week may be enough for an initial capacity decision. Separate cacheable traffic from requests that must reach the origin.
1. Measure the current baseline
Collect these values from CDN analytics, origin logs, and invoices where available:
- Viewer requests: total requests reaching the CDN.
- Edge cache hits: requests served directly by an edge location.
- Origin fetches: requests that travel from the CDN toward the origin.
- Origin response bytes: data returned by the origin to the CDN.
- Object size and TTL: especially for large files and frequently requested pages.
- Regional performance: first-byte time and total latency by user region.
- Cache purges and revalidations: events that can create bursts of origin traffic.
Do not rely on a single cache-hit percentage without checking its definition. Some dashboards count requests, while others emphasize bytes. A high request hit ratio can coexist with substantial origin bandwidth if the misses are large files. Track both request-level and byte-level behavior when possible.
2. Estimate origin fetch reduction
For a direct design, each edge may independently miss and fetch the same object. For a tiered design, several edges can share the parent cache. A practical estimate is:
Estimated origin bytes = cacheable request bytes × (1 − effective shield hit rate)
Use an effective shield hit rate for the traffic that actually reaches the tier, not the overall viewer cache-hit rate. Keep dynamic, private, bypassed, and uncacheable traffic separate. If the provider reports shield requests and shield hits, calculate:
Shield hit rate = shield hits ÷ shield requests
3. Compare total cost, not only origin bandwidth
List each architecture’s relevant cost components:
- CDN request and data-transfer charges
- Traffic from the shield or parent layer to the origin
- Origin egress or hosting transfer charges
- Additional cache-layer or regional processing charges, if applicable
- Operational cost of configuration, testing, observability, and rollback
Use the provider’s current pricing page and account-specific terms for actual rates. If a rate is unknown, leave it as a variable rather than substituting a market average. The comparison can still show the break-even point.
Inputs and assumptions
Document assumptions in a small worksheet so another engineer can reproduce the calculation. At minimum, record:
| Input | Why it matters | Recommended treatment |
|---|---|---|
| Monthly cacheable bytes | Sets the potential transfer and origin-saving base | Exclude private and bypassed responses |
| Edge hit rate | Shows how much traffic avoids deeper layers | Track by requests and bytes |
| Shield hit rate | Estimates how often a tier prevents an origin fetch | Use the provider’s shield-specific metric |
| Origin egress rate | Determines the value of reducing origin bytes | Use the current hosting or cloud rate |
| Shield transfer rate | Can offset savings if the added hop is billed | Verify direction and billing unit |
| Miss latency | Shows whether the shield improves or worsens uncached requests | Measure by region and object type |
| Purges and TTLs | Influence how long objects remain reusable | Model normal and release-day behavior |
Latency deserves separate treatment from cost. A shield region near the origin may reduce origin connection overhead, but users far from that region could see a longer miss path. Conversely, a well-placed regional parent can keep misses within a nearby network area. Compare cache hits and misses separately; a shield should not be judged only by the median latency of already-cached responses.
Protect correctness while testing. Do not broadly cache account pages, cart responses, checkout flows, or responses containing user-specific data. For WordPress and WooCommerce sites, begin with versioned static assets and carefully selected public pages. The Cloudflare caching rules guide provides a useful framework for separating HTML, assets, APIs, and WordPress paths, even when you use another CDN.
Worked examples
Example A: Direct origin delivery
Assume a site has 10 TB of monthly cacheable response bytes and currently sends 25% of those bytes toward the origin because edge misses, revalidations, and bypasses are common. The estimated origin-bound volume is:
10 TB × 25% = 2.5 TB
This is the baseline. It does not imply that all 2.5 TB is avoidable. Some requests may be intentionally dynamic or uncacheable.
Example B: Regional tier
Suppose a regional tier is expected to absorb 60% of the traffic that would otherwise travel from edge locations to the origin. The remaining estimated origin volume becomes:
2.5 TB × (1 − 60%) = 1 TB
The potential reduction is 1.5 TB, before accounting for any added tier-transfer charge. The next question is whether the tier’s cost, configuration effort, and miss latency are acceptable for that saving.
Example C: Shield region
Now assume a shield design produces an 80% shield hit rate for the same origin-bound cacheable traffic. The estimated origin volume is:
2.5 TB × (1 − 80%) = 0.5 TB
This design appears stronger for origin protection, but it is not automatically the better choice. Check whether the shield is close to the origin, whether the provider bills the additional transfer, and whether a single region creates a concentration or failover concern. Measure regional first-byte time before and after enabling it.
Decision table
| Architecture | Cache-hit behavior | Origin bandwidth | Latency profile | Complexity |
|---|---|---|---|---|
| Direct edge to origin | Each edge operates largely independently | Often highest when demand is distributed | Simple path; miss performance depends on origin distance | Lowest |
| Regional tier | Nearby edges can share parent objects | Lower when regional demand overlaps | Usually adds a parent hop on misses | Moderate |
| Origin shield | Many edges converge on a selected shield | Often most predictable for cacheable content | Depends strongly on shield and origin placement | Moderate to high |
These examples are decision aids, not forecasts. Replace the assumed rates with observed metrics, and test both normal traffic and a cache-cold event such as a deployment or large purge.
When to recalculate
Revisit the estimate whenever a material input changes. At a minimum, review it during traffic growth planning, before seasonal campaigns, after a hosting or CDN migration, and after any pricing or billing-model change. Recalculate sooner if origin bandwidth rises unexpectedly, cache-hit ratios fall, or users in one region report slower first-byte times.
Use a controlled rollout rather than changing every property at once:
- Choose a representative hostname, path group, or traffic percentage.
- Record the baseline for origin bytes, origin requests, cache-hit ratios, TTFB, error rates, and purge activity.
- Enable the tier or shield with a documented rollback step.
- Observe both warm-cache and cold-cache behavior across important regions.
- Compare cost and performance after enough traffic has passed to include ordinary misses and releases.
- Keep the change only if origin protection, cost, reliability, or operational simplicity improves without unacceptable latency or cache-safety risk.
For a broader measurement plan, see the CDN analytics guide. If the change is part of an availability strategy, pair it with a rehearsed CDN failover runbook. The best architecture is not the one with the most cache layers; it is the one whose measured benefits justify its added path, billing exposure, and operational responsibility.