Real Systems: Multi-Region Trade-offs
What you'll learn: How deploying across multiple geographic regions forces explicit CAP choices per operation type due to increased partition likelihood.
The Geography Problem
When you deploy a system across multiple regions (say, servers in New York, London, and Tokyo), the distance creates a fundamental challenge: network partitions become much more likely. A transatlantic cable failure, regional internet outage, or ISP routing issue can suddenly split your system in two—or three.
Unlike a single data center where partitions are rare emergencies, multi-region deployments face partitions frequently enough that you must design for them upfront.
Forced Trade-off Decisions
Geographic distribution means you can't pretend partitions won't happen. Instead, you must make explicit CA vs CP vs AP choices per operation type:
- User profile reads (AP): Show slightly stale data from local region rather than wait 200ms for cross-ocean consistency check
- Payment processing (CP): Block and wait for cross-region confirmation, even if it takes 500ms, to ensure no double-charges
- Content feed (AP): Display cached posts immediately; eventual consistency is acceptable
- Account balance updates (CP): Require strong consistency across all regions before confirming transaction
The key insight: different features in the same system may choose differently. Your shopping cart can be AP (merge conflicts, show best guess), while checkout is CP (never charge twice).
Multi-region forces you to answer: "When this region can't reach the others, what should this specific operation do?"
Key Takeaway: Multi-region deployments make network partitions common, not exceptional, forcing you to explicitly choose CP or AP behavior per operation type based on business requirements rather than treating consistency as a system-wide setting.