Benchmarks
Quanton delivers 2–3x better price-performance for ETL workloads compared to OSS Apache Spark.
TPC-DS results
The following results compare OSS Apache Spark vs Quanton on the TPC-DS benchmark — 99 analytical queries on Parquet.
SF-1 (1 GB dataset)
| Metric | Value |
|---|---|
| OSS Spark avg total | 156.22s |
| Quanton avg total | 103.98s |
| Avg speedup (per query) | 2.03x |
| Max speedup | 8.72x (q88) |
| Total speedup | 1.50x |
SF-10 (10 GB dataset)
| Metric | Value |
|---|---|
| OSS Spark avg total | 738.85s |
| Quanton avg total | 327.79s |
| Avg speedup (per query) | 2.71x |
| Max speedup | 7.62x (q21) |
| Total speedup | 2.25x |
Results from 3GB executor memory, 2 executors, 1 core each, on Apple M1. Gains are higher on production Spark infrastructure.
Industry benchmarks
- Onehouse Quanton vs the latest AWS EMR for Apache Spark Workloads
- Apache Iceberg on Quanton: 3x Faster Apache Spark Workloads
Run the benchmark yourself
The Quanton Operator repo includes a self-contained TPC-DS benchmark that runs locally on minikube, comparing OSS Spark vs Quanton side-by-side.
Prerequisites
- minikube with 4+ CPUs, 8+ GB memory, 50+ GB disk
- Spark Operator and Quanton Operator installed (see Local Quickstart)
Run
git clone https://github.com/onehouseinc/quanton-operator
cd quanton-operator
# Start minikube if not already running
minikube start --cpus 4 --memory 8g --disk-size 50g
# Run the benchmark (SF=1 by default)
./benchmarks/run.sh
# Custom scale factor
./benchmarks/run.sh --scale-factor 10
What it does
| Phase | Description |
|---|---|
| 0 | Builds a datagen Docker image in minikube's daemon |
| 1 | Creates a 50Gi PVC and ConfigMaps |
| 2 | Generates TPC-DS Parquet data with dsdgen |
| 3 | Runs 99 TPC-DS queries on OSS Spark |
| 4 | Runs 99 TPC-DS queries on Quanton |
| 5 | Prints a per-query comparison table with speedups |
Results are written to benchmarks/results/ as JSON.
Resource configuration
The benchmark uses: 2-core driver (4GB), 2 executors (1 core, 3GB each). Adjust in benchmarks/k8s/ YAML files for your hardware.
Using Claude Skills
If you have Claude Code installed, the /run-tpcds-benchmark skill automates the full benchmark workflow interactively. See Claude Skills for details.