PuffinDB is designed to take advantage of the very latest serverless technologies (Cf. Architecture), and what is likely to come next:
- Serverless containers (AWS Fargates) that could start in 15s instead of 60 to 90s
- Serverless functions (AWS Lambda) that could "officially" be used in a stateful manner
- Serverless functions that could be provisioned in very large numbers (hundreds of thousands)
- Serverless containers and functions that could be instantiated in larger numbers (thousands and tens of thousands respectively)
- Serverless containers and functions with 2 to 4 times more vCPUs and memory
- Serverless containers and functions with 2 to 4 times higher bandwidth/memory ratio
- Serverless containers and functions with GPU support.
- Serverless containers and functions with superchip support (NVIDIA Grace Hopper)
- Serverless Redis cluster (Amazon ElastiCache for Redis, but serverless)
- SSD tier for Object Store (Amazon S3)
- In-memory tier for Object Store (Amazon S3)
- Parquet export when using
SelectObjectContent - DuckDB embedded within Object Store (as an upgrade from
SelectObjectContent) - Object updates in place within Object Store (powered by file formats that support them, such as DuckDB's native file format)
- SQL Superchips
We look forward to working with AWS and other cloud providers to make these wishes come true.