A benchmark comparing SQLite and DuckDB on a $16.49/month Hetzner server reveals significant performance differences. DuckDB demonstrates write speeds 4x to 15x faster than SQLite and can serve dashboards with 100x the row count. Specifically, DuckDB handles a billion metric points in 10.8 GB, a scale SQLite couldn't approach.

The benchmark, implemented by integrating DuckDB as a storage engine for the Traceway binary, tested write and read performance across metrics, spans, and logs. DuckDB's columnar engine allows it to query 100x more data points without major backend changes, with read cliffs extending 100x further than SQLite's at equal or better latency.

For logs, DuckDB's write throughput is 15x higher than SQLite's, transforming a sharp performance wall into a gradual slope of polite refusals. The results suggest that self-hosting a full OpenTelemetry stack capable of handling large data volumes is feasible on a small server using DuckDB.