AWS re:Invent 2025 - Dive deep into Amazon DynamoDB (DAT435)

Dive Deep into Amazon DynamoDB

Overview of DynamoDB

• DynamoDB is a document and key-value store that provides predictable low latency at any scale
• It is used for foundational services like login, inventory management, and the control plane of most AWS services
• DynamoDB aims to provide over 99.999% availability and single-digit millisecond latency, even at massive scale

Scalability and Performance

• DynamoDB can handle over 17,000 bathtubs worth of requests per minute (equivalent to 61 seconds to fill an Olympic-sized swimming pool)
• Horizontal scaling is achieved through hash-based partitioning, where data is sorted and stored in contiguous partitions

Case Study 1: Unexpected Write Throttling

• A customer building a nationwide poll application experienced unexpected write throttling at 800 TPS, despite a simple read-write architecture
• The application used randomized IDs to ensure anonymity, but the deterministic hashing algorithm led to uneven distribution of writes across partitions
• The solution was to prefix the IDs with a few characters to leverage the "avalanche" property of the hash function and distribute writes evenly

Case Study 2: Counterintuitive Latency Patterns

• Another customer had an application with high-traffic batch workloads and low-traffic interactive workloads
• Surprisingly, the low-traffic interactive workloads experienced high and variable latency, while the high-traffic batch workloads had low, predictable latency
• This was due to the caching behavior of DynamoDB's request routers - the low-traffic workloads were not benefiting from the caches, while the high-traffic workloads were
• The solution was to route the low-traffic workloads to a smaller number of hosts to ensure they could leverage the caches

Key Takeaways

• Understanding the inner workings of the database is crucial for building effective applications
• DynamoDB's hash-based partitioning and caching mechanisms can lead to unexpected performance characteristics if not properly accounted for
• Customers should carefully consider their workload patterns and leverage DynamoDB's features to optimize for performance and cost
• Detailed monitoring and simulation can help identify and resolve performance issues in complex, real-world applications

Technical Details

• DynamoDB uses a deterministic hash function with "avalanche" properties to distribute data across partitions
• Caching is extensively used in the request routing and authentication/authorization layers to handle massive scale
• Customers can leverage prefixing or routing strategies to work around potential performance issues related to hash-based partitioning and caching

Business Impact

• The ability to handle massive scale and provide predictable low latency makes DynamoDB a critical component for mission-critical, high-availability applications
• The case studies demonstrate how DynamoDB's underlying architecture can impact real-world applications, and how customers can work around these challenges
• Understanding these performance characteristics is essential for building robust, scalable applications on top of DynamoDB