AWS re:Invent 2025 - An insider’s look into architecture choices for Amazon DynamoDB (DAT436)

Summary of AWS re:Invent 2025 Presentation: "An insider's look into architecture choices for Amazon DynamoDB"

Key Tenets of DynamoDB Design

• The DynamoDB team has four key tenets that guide their decision-making:
  1. Security: Security is non-negotiable and will never be traded off for other features.
  2. Durability: Ensuring data durability is a top priority.
  3. Availability: Maintaining high availability is critical.
  4. Predictable Low Latency: Providing predictable, single-digit millisecond latency at any scale is a core requirement.

Scaling Challenges in Distributed Systems

• Scaling a distributed database like DynamoDB comes with significant challenges:
  • Handling state and coordination in a distributed system is complex.
  • Dealing with failures, network partitions, and "split brain" scenarios requires careful design.
  • Maintaining consistency, availability, and partition tolerance (the CAP theorem) is a constant trade-off.

DynamoDB's Architectural Approach

• DynamoDB uses a distributed architecture with the following key components:
  1. Storage Node Fleet: Thousands of storage nodes organized into groups of three replicas, spread across multiple availability zones.
  2. Request Router Fleet: A fleet of request routers that handle authentication, authorization, and metadata lookups to route requests to the appropriate storage nodes.
  3. Metadata System: A system that tracks the location of data partitions and the membership of the storage node groups.

Request Routing and Latency Optimization

• To optimize latency, DynamoDB uses a multi-tiered caching approach:
  1. Request routers cache user identity and table metadata.
  2. A separate metadata cache (MDS) stores partition location information.
  3. Requests first check the caches, and if there's a miss, they make parallel requests to the MDS to hedge against latency.
• DynamoDB also uses DNS-based load balancing to route requests to the nearest available storage nodes within the same availability zone, reducing network latency.

Handling Failures and Availability Zones

• DynamoDB stripes data replicas across multiple availability zones to ensure high availability and durability.
• If an entire availability zone fails, DynamoDB can transparently route traffic to the remaining healthy zones, though this may result in higher latency.
• To maintain predictable latency, DynamoDB limits the maximum traffic that can be routed to a single availability zone, even in the event of a failure.

Partition Management and Limits

• DynamoDB partitions tables and distributes data across storage nodes to enable horizontal scaling.
• The team has introduced features like "Partition Table Warm Throughput" to help customers understand and plan for their table's capacity needs.
• DynamoDB also has various limits, such as transaction size, item size, and global secondary indexes, to ensure predictable low latency and efficient resource utilization.

Key Takeaways

• DynamoDB's design is guided by four core tenets: security, durability, availability, and predictable low latency.
• Scaling a distributed database like DynamoDB requires carefully managing state, coordination, and failures, with trade-offs between consistency, availability, and partition tolerance.
• DynamoDB uses a multi-tiered caching approach, DNS-based load balancing, and partition management techniques to optimize for low latency and high availability.
• Understanding DynamoDB's architectural choices and the reasons behind them can help customers build more effective applications on the service.