Multi-Region Systems
Two reasons, and only one of them is latency
Latency. A Sydney-to-Virginia round trip is on the order of hundreds of milliseconds, set largely by distance and routing. If users are global and dynamic work always crosses an ocean, that is a floor under those interactions.
Availability. A region is a correlated failure domain: power, network, and control-plane dependencies can fail together. Whole-region disruption is rare but possible. Whether one region is enough follows from the service's recovery objective and the actual independence of its zones and dependencies.
Both are real, and they lead to different architectures. Static and cacheable reads may be served well by a CDN and replicas; low-latency writes or regional-failure survival require more. Reach for multi-region when a measured latency or availability requirement justifies its cost.
same datacenter 0.5 ms same region, cross-AZ 1–2 ms cross-region, same ~30 ms continent cross-continent 100–200 ms Physics. Not a tuning problem.
6 components7 connections0:00
Recording…