Backend Developer interview prep for Amazon.

DynamoDB Single-Table Design collocates multiple entity types (Customers, Orders, OrderItems) into a single physical table using generic Partition Keys (PK) and Sort Keys (SK) such as `PK: CUSTOMER#123` and `SK: ORDER#456`. It is optimal for massive-scale systems requiring predictable single-digit millisecond latency at any throughput scale. For one-to-many relationships, store the parent and child entities under the same PK with differing SK prefixes, enabling retrieval of a customer profile and all their recent orders in a single, highly efficient `Query` operation rather than expensive multi-table SQL joins.

Cell-based architecture partitions a large system into multiple independent, self-contained, identical instances called 'cells'. Each cell is capable of operating completely autonomously with its own compute and data stores, capped at a maximum throughput capacity (e.g., 10,000 requests/sec). A lightweight routing layer maps customer traffic (e.g., by tenant ID or hash) to a specific cell. If a software bug, bad deployment, or data corruption occurs, the outage is strictly quarantined to that single cell (minimizing the blast radius to a small percentage of customers) rather than collapsing the entire global service.