Time-Series Storage: Design Choices That Shape Cost and Performance
InfoQ, Tuesday, May 12th, 2026
Design decisions for time-series storage affect cost and performance more than database choice itself.
This article explores fundamental design choices in time-series storage systems, demonstrating how to optimize cost and query performance using relational databases and tools like PostgreSQL. Key findings show that normalizing series identity into separate metadata tables reduces storage by 42%, while high-cardinality fields like request IDs should be excluded from series identity.
The article discusses trade-offs between flat and normalized schemas, the importance of time partitioning for efficient expiration and write distribution, and how downsampling can reduce row counts dramatically while maintaining query performance. Design decisions around compression, partitioning, and aggregation prove more impactful than the choice of database itself.