The Challenge
When we set out to build Quar.tz, we had one non-negotiable requirement: single-digit millisecond latency for any user, anywhere in the world. This article explains how we achieved it.
Edge Architecture
Traditional centralized architectures simply cannot deliver sub-10ms latency globally. The speed of light imposes hard physical limits. A request from Tokyo to a server in Virginia takes at minimum ~70ms just for the round trip.
Our solution: bring the compute to the user.
Global Edge Network
We deployed compute nodes in 47 regions across 6 continents. Each node runs our full application stack, capable of handling any request without needing to contact a central server.
Data Synchronization
The hard part isn't deploying nodes—it's keeping them synchronized. We developed a custom CRDT-based replication system that guarantees eventual consistency while allowing each node to serve requests immediately.
Results
Today, 99.9% of our requests complete in under 8ms. The remaining 0.1% are complex queries that still complete in under 50ms.