Multi-Region High Latency

Optimize cross-region latency using AWS backbone routing and data gravity principles.

Overview

Multi-region latency has a hard floor set by physics: light in fiber travels at approximately 200,000 km/s. The diagnostic question is whether your measured latency is close to that floor or far above it.

Baseline RTTs from us-east-1

Knowing the expected baseline is the first diagnostic step.

us-east-1 to us-west-2 — 70ms
us-east-1 to eu-west-1 (Ireland) — 85ms
us-east-1 to ap-northeast-1 (Tokyo) — 175ms
us-east-1 to ap-southeast-1 (Singapore) — 230ms measured, roughly 170ms theoretical minimum

> [!IMPORTANT] If measured latency matches the baseline, the only fix is to move compute or data closer to the user.

What each check tells you

Measured RTT vs the baseline table — Whether there is network optimization headroom
Route 53 routing policy — Whether users are sent to the fastest region or only the nearest one
Cross-region queries per request — Whether data gravity multiplies the latency
Cacheable vs dynamic traffic — Whether CloudFront or Global Accelerator fits better

Why the public path is slow

Public internet routing often adds 30 to 60% latency above the theoretical minimum.
Global Accelerator terminates the connection at the nearest AWS edge location, then carries traffic over AWS's private backbone.
Anycast IPs from Global Accelerator stay stable, which simplifies DNS and failover.

> [!TIP] For dynamic content that cannot be cached (API calls, streaming data), Global Accelerator typically helps more than CloudFront.

Latency-based vs geolocation routing

Geolocation — Where the client IP is physically located — French IPs always go to eu-west-3
Latency-based — RTT continuously measured from AWS edge locations — A French user may go to us-east-1 if eu-west-3 has elevated latency

> [!NOTE] For most use cases, latency-based routing produces better outcomes.

Data gravity

Data is expensive to move, in both latency and cost. Cross-region data transfer in AWS costs $0.02 to $0.09 per GB and adds a network round trip to every query.

> [!IMPORTANT] Keeping processing close to data, or replicating data to where processing happens, beats optimizing the network path.

Key concepts

AWS Global Accelerator: A networking service that routes client connections to the nearest AWS edge location, then forwards traffic over AWS private backbone. Reduces latency and provides static anycast IPs for stable endpoints.
Latency-Based Routing: Route 53 routing policy that uses measured RTT data to send each request to the AWS region that will respond fastest, not necessarily the geographically closest one.
Data Gravity: The principle that large data sets are expensive to move due to transfer costs and latency. Processing should be colocated with data rather than transferring data to processing.
Anycast: A network routing method where multiple servers share the same IP address and routing infrastructure delivers packets to the topologically nearest one — the mechanism behind Global Accelerator endpoint IPs.
Speed of Light Constraint: The theoretical minimum latency imposed by physics. Light in fiber travels ~200,000 km/s, setting a floor that no architectural change can beat — moving compute or data closer is the only solution.

When to use it

Diagnosing why a globally-distributed application has unexpectedly high latency between regions
Explaining the trade-offs between Global Accelerator and CloudFront in system design interviews
Designing multi-region architectures that minimize cross-region data transfer

Interview questions

Your API is deployed in us-east-1 and serves users in Southeast Asia who report 600ms average response times. The application processing itself takes under 50ms. What options do you have, and what are the trade-offs?

Multi-Region High Latency

Overview

Overview

Baseline RTTs from us-east-1

What each check tells you

Why the public path is slow

Latency-based vs geolocation routing

Data gravity

Key concepts

When to use it

Interview questions

More Troubleshooting topics