Direct comparison of execution time for Parquet-only TPC-H queries. Strake leverages DataFusion's multi-threaded execution for heavy aggregations.
| Query | Strake (Release) | DuckDB (Release) | Factor |
|---|---|---|---|
| Q1 Aggregation Heavy |
0.2788s | 0.5724s | 2.05x Faster |
| Q6 Scan & Filter |
0.1932s | 0.0660s | DuckDB Leads |
Strake uniquely enables joins across disparate sources (e.g., PostgreSQL + Parquet) without data movement overhead.
| Query | Description | Execution Time | Data Sources |
|---|---|---|---|
| Q3 Shipping Priority |
3-way join with filter & sort | 1.18s | Postgres + Parquet |
| Q10 Returned Items |
Top 20 analysis | 0.28s | Postgres + Parquet |
Strake's DataFusion core scales linearly with core count. This explains the 2x lead in Q1 aggregation, even on medium datasets.
The sub-second latency for Q6 remains competitive while providing governance and network interface layers that DuckDB lacks.