Skip to main content

Aurora

Income note

Amazon Aurora is a fully managed, high-performance relational database engine compatible with MySQL and PostgreSQL. It is part of Amazon Web Services (AWS) and is designed to combine the speed and reliability of high-end commercial databases with the simplicity and cost-effectiveness of open-source databases.

Aurora is built for the cloud, providing superior performance, scalability, and availability while reducing administrative overhead. It is optimized for workloads requiring high throughput, low latency, and enterprise-grade reliability.

Key Features of Amazon Aurora​

High Performance​

  • 5x Faster than MySQL: Aurora delivers up to five times the throughput of standard MySQL running on the same hardware.
  • 3x Faster than PostgreSQL: It provides up to three times the performance of standard PostgreSQL.
  • Low-Latency Read Replicas: Aurora supports low-latency read replicas for scaling read-heavy workloads.

Scalability​

  • Auto-Scaling Storage: Aurora automatically scales storage from 10 GiB up to 128 TiB without downtime.
  • Compute Scaling: Database instances can be scaled up or down with minimal downtime.
  • Read Replicas: Supports up to 15 read replicas (compared to MySQL’s 5) for horizontal read scaling.

High Availability & Durability​

  • Multi-AZ Deployments: Aurora replicates data across multiple Availability Zones (AZs) for fault tolerance.
  • Self-Healing Storage: Automatically detects and repairs disk failures.
  • Continuous Backups: Point-in-time recovery and automated backups with no performance impact.

Fully Managed​

  • Automated Patching & Updates: AWS handles database patching and minor version upgrades.
  • Backup & Restore: Automated backups with retention periods up to 35 days.
  • Monitoring & Alerts: Integrated with Amazon CloudWatch for performance metrics and alarms.

Security & Compliance​

  • Encryption at Rest & in Transit: Supports AES-256 encryption for data at rest and SSL/TLS for data in transit.
  • Network Isolation: Works with Amazon VPC for secure network configurations.
  • IAM Integration: Fine-grained access control using AWS Identity and Access Management (IAM).
  • Compliance Certifications: Compliant with SOC, PCI-DSS, HIPAA, ISO, and GDPR.

Cost-Effective​

  • Pay-as-You-Go Pricing: No upfront costs; pay only for the compute and storage used.
  • No License Fees: Unlike traditional enterprise databases, Aurora has no licensing costs.
  • Serverless Option: Aurora Serverless automatically adjusts capacity based on demand, reducing costs for intermittent workloads.

Aurora Architecture Internals​

Amazon Aurora reimagines traditional database architecture by decoupling storage and compute while leveraging a distributed, fault-tolerant design. Below is a detailed breakdown of Aurora’s internal architecture, replication model, durability mechanisms, and engine-level differences compared to standard MySQL and PostgreSQL.

Aurora’s Distributed & Decoupled Storage-Compute Architecture​

Separation of Compute and Storage​

  • Unlike traditional databases where storage is tightly coupled with compute (local disks or attached EBS volumes), Aurora decouples storage from compute.
  • Compute Layer: Handles query processing, transactions, and caching (using DB instances).
  • Storage Layer: A distributed, auto-scaling, fault-tolerant storage system that spans multiple Availability Zones (AZs).

Key Benefits of Decoupled Architecture​

  • Independent Scaling: Compute (CPU/RAM) and storage can scale independently.
  • High Availability: Storage is replicated across AZs, so compute failures do not affect data durability.
  • Faster Recovery: Since storage is separate, a failed DB instance can be quickly replaced without data movement.

Quorum-Based Replication (4 out of 6 Replicas)​

How Aurora Ensures Durability​

  • Aurora’s storage is distributed across 6 storage nodes (2 per AZ in a 3-AZ setup).
  • Data is written using a quorum model: 4 out of 6 replicas must acknowledge writes before a transaction is considered durable.
    • This ensures tolerance for up to 2 storage node failures without data loss.
    • Reads require 3 out of 6 replicas to maintain consistency.

Advantages Over Traditional Replication​

FeatureAurora (Quorum Model)Traditional MySQL/PostgreSQL (Synchronous Replication)
Write Durability4/6 replicas (strong consistency)Typically 2 replicas (primary + standby)
Failure ToleranceSurvives 2 storage node failuresUsually fails if primary & standby both crash
Performance ImpactLower latency (parallel writes)Higher latency (sequential replication)

Redo Log Buffering & Durability Guarantees​

3.1 Log-Structured Storage​

  • Aurora does not write data pages directly to disk. Instead, it only ships redo logs to storage nodes.
  • The storage layer reconstructs data pages from logs in the background, reducing I/O overhead.

3.2 Redo Log Optimization​

  • Minimizes Write Amplification: Only log records are sent over the network, not entire pages.
  • Asynchronous Page Materialization: Storage nodes apply logs in the background to reconstruct pages.
  • Durability Without fsync() Delays: Aurora avoids waiting for disk flushes (fsync) by relying on quorum commits.

3.3 Comparison with Traditional Databases​

MechanismAuroraStandard MySQL/PostgreSQL
Primary Write MechanismShips redo logs onlyWrites full data pages + logs
Durability LatencyFaster (quorum-based)Slower (depends on disk sync)
Recovery TimeNear-instant (logs reconstruct state)Slower (requires replaying logs)

4. Aurora MySQL vs. Aurora PostgreSQL Internals​

4.1 Aurora MySQL​

  • Based on MySQL 5.7/8.0 but with significant optimizations:
    • No InnoDB Double-Write Buffer: Aurora’s storage layer eliminates the need for this by ensuring crash-safe page writes.
    • Parallel Query Execution: Improves analytical query performance.
    • Optimized Buffer Pool Management: Reduced contention for memory access.

4.2 Aurora PostgreSQL​

  • Based on PostgreSQL 11+ with AWS enhancements:
    • Native PostgreSQL Replication Replaced: Uses Aurora’s distributed storage instead of PostgreSQL’s WAL shipping.
    • Optimized Vacuuming: Reduced bloat impact due to Aurora’s storage layer.
    • Better Read Scalability: Supports up to 15 read replicas (vs PostgreSQL’s ~10).

4.3 Key Differences Between Aurora MySQL & PostgreSQL​

FeatureAurora MySQLAurora PostgreSQL
Underlying EngineMySQL 5.7/8.0PostgreSQL 11+
Replication ModelAurora storage layer (not MySQL binlog)Aurora storage layer (not PostgreSQL WAL)
Buffer PoolOptimized InnoDB buffer poolShared buffers + Aurora optimizations
Parallel QuerySupportedSupported (but different implementation)

5. How Aurora Differs from Standard MySQL/PostgreSQL at the Engine Level​

5.1 Storage Engine Modifications​

  • No Direct Disk Writes: Aurora bypasses the local filesystem and sends logs directly to the distributed storage layer.
  • No Double-Write Buffer (MySQL): Not needed due to Aurora’s crash-safe storage.
  • No WAL Archiving (PostgreSQL): Aurora handles log shipping internally.

5.2 Networking & I/O Optimizations​

  • Bypasses OS Page Cache: Aurora uses a user-space network stack for lower-latency log shipping.
  • Zero-Copy Log Shipping: Reduces CPU overhead by avoiding unnecessary data copies.

5.3 Recovery & Crash Consistency​

  • Instant Crash Recovery: Since storage nodes have quorum-committed logs, recovery is near-instant.
  • No Need for Crash Recovery (PostgreSQL): Unlike standard PostgreSQL, Aurora does not require lengthy pg_wal replay.

5.4 Performance Enhancements​

  • Read Scaling: Up to 15 low-lag replicas (vs MySQL’s 5, PostgreSQL’s ~10).
  • Write Scaling (Multi-Master): Aurora PostgreSQL supports multi-master clustering for high write throughput.