Why Singapore was 380ms slower than London

Why Singapore was 380ms slower than London

A B2B analytics dashboard felt instant in London and sluggish in Singapore — users complained, but synthetic checks looked fine. We diagnosed it region by region.

— Ran RUM split by continent: London p75 was 90ms, Singapore p75 was 470ms
— Traced it: the nearest POP was serving but every API call still went MISS to a US-East origin
— The dashboard's JSON wasn't cacheable — Cache-Control: private on every response
— Marked read-only reference endpoints public, max-age=120 and cached them at the edge
— Left user-specific calls private but added an edge compute layer to assemble the page

Singapore p75 fell from 470ms to 140ms. The win wasn't moving the origin — it was making the heavy, shared API responses cacheable near the reader. Always split latency by region before optimizing; a global average hides the one continent that's hurting.

The number that mattered: 330ms cut from every Singapore page load.