S3 vs. DynamoDB

Comparing S3 with DynamoDB #

S3 is designed for throughput, not necessarily predictable (or very low) latency. It can easily deal with bursts in traffic requests, especially if the requests are for different items.

DynamoDB is designed for low latency and sustained usage patterns. If the average item is relatively small, especially if items are less than 4KB, DynamoDB is significantly faster than S3 for individual operations. Although DynamoDB can scale on-demand, it does not do that as quickly as S3. If there are sudden bursts of traffic, requests to DynamoDB may end up throttled for a while.

S3 operations generally work on entire items. Atomic batch operations on groups of objects are not possible, and it’s difficult to work with parts of an individual object. There are some exceptions to this, such as retrieving byte ranges from an object, but appending content to a single item from multiple sources concurrently is not easy.

DynamoDB works with structured documents, so its smallest atom of operation is a property inside an item. You can, of course, store binary unstructured information to DynamoDB, but that’s not really the key use case. For structured documents, multiple writers can concurrently modify the properties of the same item, or even append to the same array. DynamoDB can ...