Skip to content

PATTERN Cited by 1 source

Primary-fallback PoP separation

Pattern

When selecting primary and fallback upper tiers for a cache hierarchy, always place them on different physical points of presence (PoPs). This ensures that losing a single PoP (due to fiber cut, power loss, maintenance) cannot simultaneously eliminate both the primary and fallback cache paths to origin.

Rationale

If primary and fallback tiers were co-located in the same physical facility: - A single PoP outage would remove both tiers. - All edge POPs dependent on that region would need to fall further back (to origin directly, or to a tertiary tier if one exists). - The blast radius of one facility failure expands to affect all cache miss traffic for the associated origins.

By enforcing PoP diversity: - Primary failure → failback to a geographically separate PoP (still one hop from origin). - Single-facility events never take out both paths. - Analogous to availability-zone separation in cloud architecture.

Implementation Note

This constraint is applied after the tier-selection algorithm runs: once the primary upper tier is elected (by latency or weighted voting), the fallback is selected from the remaining candidates with the additional constraint of being on a distinct PoP.

Seen in

Last updated · 575 distilled / 1,757 read