Aurora DB

Aurora decouple compute from storage

Aurora Storage:

• Storage layer is ALWAYS replicated across 3 AZs

• This is automatic and happens regardless of how many DB instances you have

• Data is replicated 6 ways (2 copies per AZ)

Aurora DB Instances:

• Can be deployed in 1, 2, or 3 AZs

• Minimum recommended for high availability is 2 AZs

  • 1 primary instance in one AZ

  • At least 1 replica in another AZ

• You can choose to deploy instances in 3 AZs for additional availability

Core Characteristics

Fully managed RDS service supporting MySQL and PostgreSQL databases

• Built-in multi-AZ availability by design

• Automatic storage scaling managed by AWS

• Easy multi-region replication capabilities

Architecture and High Availability

Instance Configuration

• Primary instance + up to 15 read replicas

• Instances handle compute and networking

• Data stores distributed across availability zones

• Independent scaling of storage from instances

Data Management

• Primary instance handles all write traffic

• Data replication typically available within 100ms

• Both primary and read replicas can handle read traffic

• Automatic failover handling without endpoint changes

Endpoints

• Cluster endpoint: Routes to primary instance for writes

• Reader endpoint: Routes through ELB to distribute read traffic

• Optional custom instance and group endpoints available

Global Databases

Configuration

• Single primary region with up to 5 secondary regions

• Storage-level replication for minimal latency

• Cross-region data typically available in under 1 second

• Secondary regions can be promoted during outages

Aurora Serverless

Key Features

• Automatic instance scaling for variable workloads

• Measured in Aurora Capacity Units (ACUs)

• 1 ACU ≈ 2GB memory + proportional CPU and networking

• Scaling range: 0.5 to 128 ACUs

• Storage automatically scales from 10 GiB to 128 TiB

• Storage scaling is independent of ACU scaling

• Stepwise scale-down approach similar to EC2 auto-scaling

Use Cases

• MySQL and PostgreSQL workloads requiring high availability

• Applications needing resilient database infrastructure

• Variable/spiky workload patterns (using Serverless)

• Scenarios requiring minimal operational overhead

• Read-heavy applications benefiting from replica distribution

Performance Characteristics

• Fast replication (sub-100ms within region)

• Sub-second replication for global databases

• Elastic scaling for both storage and compute

• Distributed read workload capabilities