TPC-H benchmark
TrinoX vs. Trino-Java
End-to-end TPC-H Q1–Q22 query latency in milliseconds (lower is better) on identical hardware and dataset layout. Speedup = Trino-Java ÷ TrinoX.
Latency — SF1
Speedup — SF1
Geomean speedup by scale factor
| Scale factor | TrinoX / Trino-Java |
|---|
Full results (all scale factors)
Methodology
TrinoX
Standalone Rust engine
Native vectorized execution over Parquet-backed TPC-H tables. Same Trino-compatible HTTP surface; queries run through the in-tree planner and executor (single-node or distributed).
Trino-Java
JVM reference
Stock Trino coordinator/worker on the same cluster shape and data layout. Numbers are wall-clock query latency from a controlled benchmark harness — not extrapolated from micro-benchmarks.