Amazon Aurora’s distributed, log-structured storage architecture decoupling compute and storage, using six-way replication, quorum-based writes, striping, auto-growing volumes, continuous S3 backups, and automated repair for durability.
Welcome back. In this lesson we’ll examine how Amazon Aurora implements distributed storage and why that design improves scalability, availability, and durability for managed relational databases.
Traditional single-instance databases combine compute and storage on the same server. That approach ties two different lifecycles together:
Compute is often ephemeral and scaled independently (you may replace or resize instances).
Storage must remain durable and highly available across failures and AZ outages.
Decoupling compute from storage gives independent scaling, faster failover, and more cost-effective lifecycle management for each layer.
This lesson explains Aurora’s distributed, log-structured storage architecture and how it uses replication and quorum rules to provide high durability and low-latency access.
Aurora separates the storage layer from compute and implements a purpose-built, log-structured distributed storage system optimized for cloud scale and fast recovery.Key characteristics at a glance:
Characteristic
What it means
Log-structured storage
Changes are written as an append-only log, enabling fast sequential I/O and efficient crash recovery.
Striping across SSDs
Cluster volumes are striped across many storage nodes that use locally attached SSDs for low-latency I/O.
Continuous backups to S3
Incremental backups are continuously streamed to Amazon S3 for durable, off-cluster snapshots.
Multi-AZ distribution
Data is replicated across multiple Availability Zones to tolerate AZ failures.
10 GB protection segments
The cluster volume is segmented into 10 GB units (protection groups) for replication and repair.
Auto-growing volumes
Cluster volumes expand automatically up to 128 TB as needed.
When the database writes data, Aurora divides the cluster volume into 10 GB segments (protection groups). Each segment is replicated six times across the region — typically implemented as two copies per Availability Zone across three AZs (2 × 3 = 6).This six-way replication across multiple AZs protects against individual storage node failures and AZ-level outages while enabling parallel I/O and fast rebuilds of damaged copies.
Aurora uses quorum-based rules to balance durability with low-latency I/O. The cluster maintains six copies of each 10 GB segment and applies the following rules:
Operation
Quorum required
Purpose
Durable write acknowledgement
4 of 6 replicas
Guarantees that a majority of replicas have persisted the change before returning success.
Read servicing
Any healthy replica (often local/fast)
Reads may be served from any replica; specific recovery or consistency operations can require more replicas.
Minimal recovery/consistency check
3 of 6 replicas
Used in some repair and recovery workflows to determine segment health.
This design allows Aurora to acknowledge writes quickly (once 4 replicas persist the log record), while still tolerating up to two replica failures without losing durability guarantees.
Aurora’s storage layer handles replication, integrity checking, and automated repair:
Replication: Updates are propagated to the storage nodes that host the segment copies.
Integrity: Checksums and metadata detect corruption or mismatch among copies.
Repair: If a copy is corrupt or a node fails, Aurora copies data from healthy replicas to rebuild the missing/corrupt copy automatically.
Under the hood, Aurora uses background protocols (often described as gossip-like or membership/heartbeat mechanisms) to monitor node health, coordinate repair, and restore full redundancy without manual intervention.
Replication traffic between storage nodes travels over the AWS network and is protected by internal integrity checks.
Client connections to Aurora support TLS for transport-level encryption.
When you enable encryption for an Aurora cluster, data at rest is encrypted across the storage layer, and internal replication respects those protections.
Amazon Aurora’s distributed storage is a core contributor to its performance and durability profile. The key elements are:
Log-structured append-only storage
10 GB protection segments
Six-way replication across multiple AZs (typically 2 copies per AZ × 3 AZs)
Quorum rules (4/6 writes for durability; reads from healthy replicas)
Continuous backups to Amazon S3
Automated detection and repair of corrupted or missing copies
These features together deliver a resilient, scalable storage substrate for managed relational databases that supports large, auto-growing cluster volumes and fast, reliable operation.
