Skip to content

Latest commit

 

History

History
112 lines (94 loc) · 4.07 KB

File metadata and controls

112 lines (94 loc) · 4.07 KB

TPC-H — DeltaForge benchmark

22 canonical TPC-H queries on the 8-table schema, SF=1.

One page per benchmark; full methodology in methodology.md. Other benchmarks: TPC-H · TPC-DS · SSB · JOB · Index

Completion summary

engine queries completed warm-median (ms) first error
DeltaForge 22 / 22 255.44 no failures
DuckDB 22 / 22 173.12 no failures
Spark (default) 22 / 22 1,477.74 no failures
Spark (tuned) 21 / 22 1,527.79 Py4JJavaError('An error occurred while calling o2096.collectToPython.\n', JavaObject id=o2097)

Warm-median execution time (ms)

The headline number per (engine, query) is the median of 9 warm runs. df numbers are server-reported SHOW STATS ACTUAL.total_time_ms; DuckDB and Spark numbers are wall around the SELECT (views pre-registered in untimed setup, attach cost excluded for every engine).

query DeltaForge DuckDB Spark (default) Spark (tuned)
q01 479.20 189.18 2,750.44 2,593.72
q02 167.84 147.56 1,714.15 1,291.67
q03 288.96 180.48 1,507.46 1,744.64
q04 229.32 140.21 1,175.88 1,527.79
q05 383.61 188.16 2,346.87 -
q06 152.67 86.46 333.61 316.18
q07 286.60 202.08 1,798.81 1,673.40
q08 297.87 241.58 2,038.35 1,722.26
q09 382.68 298.78 2,414.01 1,844.77
q10 317.54 271.79 1,962.45 1,855.07
q11 101.93 109.29 1,306.33 1,213.71
q12 287.02 132.72 988.73 1,320.47
q13 188.84 173.23 1,175.60 1,723.19
q14 153.66 139.11 595.11 589.17
q15 239.45 108.94 1,352.43 1,153.82
q16 162.56 154.02 1,244.48 1,209.29
q17 271.43 173.01 1,602.09 2,506.71
q18 617.00 219.62 2,658.12 2,649.64
q19 222.93 170.38 663.65 684.83
q20 221.44 187.50 1,448.02 1,203.50
q21 526.15 381.75 3,244.01 2,965.08
q22 108.66 130.17 1,415.45 1,439.54
median (across completed queries) 255.44 173.12 1,477.74 1,527.79

Speedup vs DeltaForge (higher = faster than df)

query DuckDB Spark (default) Spark (tuned)
q01 2.53x 0.17x 0.18x
q02 1.14x 0.10x 0.13x
q03 1.60x 0.19x 0.17x
q04 1.64x 0.20x 0.15x
q05 2.04x 0.16x -
q06 1.77x 0.46x 0.48x
q07 1.42x 0.16x 0.17x
q08 1.23x 0.15x 0.17x
q09 1.28x 0.16x 0.21x
q10 1.17x 0.16x 0.17x
q11 0.93x 0.08x 0.08x
q12 2.16x 0.29x 0.22x
q13 1.09x 0.16x 0.11x
q14 1.10x 0.26x 0.26x
q15 2.20x 0.18x 0.21x
q16 1.06x 0.13x 0.13x
q17 1.57x 0.17x 0.11x
q18 2.81x 0.23x 0.23x
q19 1.31x 0.34x 0.33x
q20 1.18x 0.15x 0.18x
q21 1.38x 0.16x 0.18x
q22 0.83x 0.08x 0.08x

Host

CPU 85 (18 physical / 36 threads)
Memory ? GiB total
Cgroup ? CPUs, ? MiB
Disk ? on ? — read ? MB/s, write ? MB/s
Virt {'container_hint': False, 'hypervisor_flag': True, 'wsl2': True}
Run started 2026-05-18T16:05:14.494487+00:00
Run id 20260518T131418Z-8ae0556ef6b8-publish-tpch

Engine versions

engine version
DeltaForge ?
DuckDB ?
Spark (default) ?
Spark (tuned) ?

Reproducing this table

docker compose exec bench python data_gen/generate_tpch_delta.py --scale 1
docker compose exec bench python bench_runner.py \
    --scale 1 --engines df,duckdb,spark-default,spark-tuned \
    --workloads tpch_read_delta

Raw per-step records: 20260518T131418Z-8ae0556ef6b8-publish-tpch/raw/. Manifest: 20260518T131418Z-8ae0556ef6b8-publish-tpch/manifest.json.


Next: TPC-DS · Methodology · Index