AWS re:Invent 2025 - Data modeling core concepts for Amazon DynamoDB (DAT311)

Data Modeling Core Concepts for Amazon DynamoDB

Dynamo as a Phone Book Analogy

• Dynamo is like a physical phone book, with a partition key (city) and a sort key (last name)
• This allows querying by specific PKSK combinations, as well as operations like "starts with" on the sort key
• Limitations of the phone book model, like not being able to query by other attributes, are addressed through indexes

Partition Keys and Data Distribution

• Partition keys are hashed to ensure even data distribution across partitions
• Partitions have limits (3,000 RCU, 1,000 WCU) and will split when they reach capacity
• Partition key design is crucial for even distribution and scaling

Modeling Techniques

• Partition key should be descriptive but allow for even hashing (e.g. "cust_id_hash123")
• Sort keys can be used to model hierarchical or time-based data (e.g. "order_id", "timestamp")
• Single table design is common, storing multiple entity types in the same table using the sort key prefix

Scaling and Performance

• Compression can reduce storage and write costs by storing data in a compressed format
• Soft deletes and background cleanup can handle large-scale deletes without impacting user experience
• Sharding partition keys can distribute high-write workloads across multiple partitions

Integrating with Amazon OpenSearch

• DynamoDB does not have built-in text search capabilities
• Integrating with Amazon OpenSearch allows advanced indexing and querying for text-based search

Point-in-Time Recovery and Incremental Exports

• Point-in-time recovery allows restoring the database to a previous state in case of application-level corruption
• Incremental exports to S3 provide a cost-effective way to recover from data issues, compared to full table restores

Key Takeaways

• Partition key design is crucial for even data distribution and scaling
• Indexes, compression, and sharding can optimize performance and cost
• Integrating with Amazon OpenSearch enables advanced text-based search
• Point-in-time recovery and incremental exports provide robust data protection and recovery options