When first learning DynamoDB, you hear "you must choose the partition key well," but what "choosing well" means mathematically is rarely taught well. In this day, we understand why partition key design is a mathematical problem, how LSI and GSI are stored differently internally, and why Single-Table Design has become the standard pattern for DynamoDB where dozens of microservices previously used separate tables.
Mathematically defining a "good" partition key choice: reads/writes load must be evenly distributed across all partitions.
The metrics measuring this even distribution are Cardinality and Skewness.