Multi-AZ: deploy resources trên nhiều Availability Zones trong cùng region (thường 2-3 AZ cách nhau vài chục km, latency 1-2ms); AWS RDS Multi-AZ tự động failover khi primary AZ fail (<30s), ALB tự động route traffic qua AZ available, Auto Scaling Group spread instances qua AZs; chi phí thêm ~2x cho redundant resources nhưng vẫn trong region; bảo vệ khỏi AZ failure (datacenter fire, power outage, connectivity).
Multi-Region: deploy trên nhiều AWS regions (cách nhau hàng nghìn km, latency 100-300ms inter-region); bảo vệ khỏi region-level disaster (cực hiếm nhưng xảy ra — us-east-1 có vài incident lớn), giảm latency cho global users, đáp ứng data residency requirements (GDPR — EU data phải ở EU).
Đánh đổi multi-region: phức tạp hơn nhiều (data replication lag, eventual consistency, split-brain), chi phí tăng 2-3x (replicate storage, data transfer cross-region $0.02/GB), operational overhead cao. Strategies: Active-Passive (traffic dồn vào primary region, secondary là warm standby chỉ dùng khi failover), Active-Active (traffic split đều hai region, cả hai phục vụ user).
Services hỗ trợ multi-region: Aurora Global Database, DynamoDB Global Tables, S3 Cross-Region Replication, Route 53 latency/failover routing, CloudFront. RTO/RPO quyết định strategy: multi-AZ đủ cho RTO<5 phút, multi-region cần khi RTO/RPO gần 0.
Multi-AZ: deploy resources across multiple Availability Zones within the same region (typically 2-3 AZs tens of kilometers apart, with 1-2ms latency); AWS RDS Multi-AZ automatically fails over when the primary AZ fails (<30s), ALB automatically routes traffic to available AZs, and Auto Scaling Groups spread instances across AZs; cost increases ~2x for redundant resources but remains within the region; protects against AZ failures (datacenter fire, power outage, connectivity loss).
Multi-Region: deploy across multiple AWS regions (thousands of kilometers apart, 100-300ms inter-region latency); protects against region-level disasters (extremely rare but have occurred — us-east-1 has had several major incidents), reduces latency for global users, and meets data residency requirements (GDPR — EU data must stay in the EU).
Multi-region trade-offs: significantly more complex (data replication lag, eventual consistency, split-brain scenarios), 2-3x higher costs (replicating storage, cross-region data transfer at $0.02/GB), and high operational overhead. Strategies: Active-Passive (traffic goes to the primary region; secondary is a warm standby used only during failover), Active-Active (traffic is split between both regions; both serve users simultaneously).
Services supporting multi-region: Aurora Global Database, DynamoDB Global Tables, S3 Cross-Region Replication, Route 53 latency/failover routing, and CloudFront. RTO/RPO determines the strategy: multi-AZ is sufficient for RTO < 5 minutes; multi-region is needed when RTO/RPO must be near zero.