Real-World Data Benchmarks
Benchmarks built on public real-world datasets rather than synthetic TPC or vendor data.
Why Real-World Data?
Synthetic benchmarks like TPC-H generate data that is statistically clean: uniform distributions, predictable cardinalities, well-behaved joins. Real-world data is messier - skewed, sparse, seasonal, and full of edge cases. BenchBox’s real-world benchmarks capture these characteristics using authoritative public datasets.
What real-world data exposes that synthetic data hides
- Temporal skew - weekday vs. weekend, rush hour vs. overnight, holiday effects
- Geographic skew - airport hubs, city centers, regional carriers
- Categorical imbalance - a handful of carriers / zones / host classes dominate
- Sparsity - not every combination of dimensions exists in the data
- Dirty data - nulls, rare categories, outliers that force defensive SQL
These characteristics drive optimizer behavior in ways TPC workloads do not - cardinality estimation, join ordering, partition pruning, and predicate pushdown all behave differently.
Real-World Benchmarks in BenchBox
| Benchmark | Data Source | Focus |
|---|---|---|
| NYC Taxi | NYC TLC trip records | Geospatial + temporal OLAP, taxi zones dimension, 25 queries |
| Flight Data | US BTS On-Time Performance | Aviation analytics - delays, routes, carriers, seasonality, 20 queries |
When to Use
- NYC Taxi - geospatial joins, pickup/dropoff zone rollups, fare analytics
- Flight Data - test temporal and categorical aggregation on a 20+ year government dataset
Included Benchmarks
See Also
- Time-Series Benchmarks - Time-series workloads (TSBS DevOps)
- Industry Benchmarks - Vendor / practitioner benchmarks (ClickBench, H2O, CoffeeShop)
- TPC Standards - Synthetic standard benchmarks