CDNs: edge hits and origin protection

An edge cache serves reusable responses near readers. A hit avoids the origin round trip. A miss still needs an upstream fetch, so proximity alone does not guarantee fast responses or origin protection.

A cache key identifies a representation. Omitting a meaningful input mixes responses; including irrelevant inputs fragments hits. Respect Vary and the CDN policy. Private or personalized responses generally bypass shared caching; adding a user identifier alone is not a complete privacy policy.

TTL controls freshness, not physical deletion. s-maxage targets shared caches; max-age also affects browser caches. Once stale, a response may be revalidated rather than fully downloaded. Stale-serving directives can extend reuse, so TTL alone is not an unconditional end-to-end staleness guarantee.

An origin shield adds a shared cache tier between edges and origin. Repeated edge misses can become shield hits. It adds a hop on a shield miss and another operational dependency. Simultaneous cold misses need request coalescing to avoid duplicate origin fetches.

Versioned URLs keep immutable assets reusable while new references select new content. Purges remove cached entries for an existing URL, with provider-specific scope and delay. CDN invalidation does not erase browser copies. Neither method prevents a post-purge surge of misses without origin capacity or coalescing.

Origin rate = incoming rate ? (1 ? edge hit ratio) ? (1 ? shield hit ratio on edge misses). Here 10,000 ? 0.1 ? 0.2 = 200 requests/s: 98% request offload. Byte offload can differ when misses are larger than hits.

Request hit ratio counts requests. Byte hit ratio weights response sizes. Nine tiny hits and one huge miss give 90% request hits but only 9 / 1,009 ? 0.9% byte hits. Track both when origin bandwidth is the bottleneck.