← Blog

EngineeringCaching

Tiered edge caching: fewer origin fetches, faster tails

May 8, 2026 · Platform Team

A cache miss at the edge used to mean a request all the way back to your origin. For popular-but-not-hot objects, that added avoidable latency and origin load. This week we shipped a regional mid-tier that sits between the edge and your origin.

How it works

When an edge PoP misses, it now checks a regional tier before reaching out to your origin. The regional tier aggregates demand from every PoP in the area, so a single origin fetch warms many edges at once.

Client → Edge PoP → Regional tier → Origin
                 ↘ (hit)        ↘ (hit)

Results

Across early-access zones we measured a 35–55% reduction in origin requests and a clear improvement at the tail — p99 dropped because fewer requests pay the full origin round trip. Cache invalidation still propagates globally via the purge API, including the new tier.

Tiering is on by default for all zones; there’s nothing to enable. You can watch the effect in your dashboard under Cache → Origin offload.