Skip to content

Platform Documentation

Documentation for database platforms supported by BenchBox.

Platform Guides

DataFrame Platforms (Native API)

BenchBox supports benchmarking DataFrame libraries using their native APIs instead of SQL. This enables direct performance comparison between SQL and DataFrame paradigms on identical workloads.

Available DataFrame Platforms

PlatformCLI NameFamilyStatusDocumentation
Polarspolars-dfExpressionProduction-readyPolars DataFrame
Pandaspandas-dfPandasProduction-readyPandas DataFrame
Daskdask-dfPandasProduction-readyDask DataFrame
cuDFcudf-dfPandasProduction-readycuDF DataFrame
PySparkpyspark-dfExpressionProduction-readyPySpark DataFrame
LakeSaillakesail-dfExpressionProduction-readyLakeSail DataFrame
DataFusiondatafusion-dfExpressionProduction-readyDataFusion DataFrame
Databricksdatabricks-dfExpressionProduction-readyDatabricks DataFrame
Terminal window
benchbox run --platform polars-df --benchmark tpch --scale 0.1
benchbox run --platform pandas-df --benchmark tpch --scale 0.1
benchbox run --platform dask-df --benchmark tpch --scale 0.1
benchbox run --platform cudf-df --benchmark tpch --scale 0.1
benchbox run --platform pyspark-df --benchmark tpch --scale 0.1
benchbox run --platform lakesail-df --benchmark tpch --scale 0.1
benchbox run --platform datafusion-df --benchmark tpch --scale 0.1
benchbox run --platform databricks-df --benchmark tpch --scale 0.1
benchbox run --platform polars --benchmark tpch --scale 0.1
benchbox run --platform polars-df --benchmark tpch --scale 0.1

Polars is the recommended starting point because it is fast. Pandas offers a familiar API. Dask is distributed. cuDF runs on GPUs and works on Linux only. PySpark fits the Spark ecosystem. LakeSail runs on Sail, a fast Spark-compatible engine. Databricks uses Databricks Connect.

The last two commands compare SQL and DataFrame execution on the same workload: --platform polars is SQL mode and --platform polars-df is DataFrame mode.

SQL Platforms

Core Local Databases

These platforms are included in the base BenchBox installation with no additional dependencies:

  • DuckDB - Embedded analytical database (default local platform)
  • SQLite - Embedded row-store database for lightweight testing

Local/Embedded Analytics Engines

Traditional Relational Databases

  • PostgreSQL - Open-source relational database with TimescaleDB support
  • CedarDB - High-performance HTAP engine over PostgreSQL wire protocol (formerly Umbra)
  • GizmoSQL - DuckDB over Arrow Flight SQL through the official ADBC driver (experimental)

PostgreSQL Extensions

  • pg_duckdb - DuckDB-powered Postgres extension for vectorized OLAP
  • pg_mooncake - Columnar storage for Postgres with Iceberg and Delta Lake support
  • paradedb - Hybrid search and analytics Postgres distribution (BM25 via pg_search)
  • citus - Distributed Postgres extension with sharded tables

Self-Hosted OLAP Databases

  • StarRocks - MPP columnar OLAP database (MySQL protocol + Stream Load)
  • Apache Doris - MPP real-time analytical database (MySQL protocol + Stream Load)
  • SingleStore - Distributed SQL with columnstore analytics (MySQL protocol + LOAD DATA)
  • QuestDB - Time-series database (PostgreSQL wire protocol + REST API)

Distributed SQL Engines

  • PrestoDB - Distributed SQL query engine (Facebook fork)
  • Trino - Distributed SQL query engine (community fork, formerly PrestoSQL)
  • Apache Spark - Unified analytics engine for large-scale data processing
  • LakeSail Sail - High-performance Spark-compatible engine (Spark Connect)
  • Apache Gluten + Velox - Spark plugin that offloads physical operators to a vectorized C++ engine (Linux-only, Docker on macOS/Windows)

Cloud Data Warehouses

GPU-Accelerated Platforms

  • CUDF - NVIDIA RAPIDS GPU-accelerated DataFrames

Platform Categories

By Installation Complexity

Zero Config (included in base install):

  • DuckDB, SQLite

Single Extra (one pip install command):

  • DataFusion, Polars, PostgreSQL, ClickHouse

Cloud SDK Required (authentication setup needed):

  • Databricks SQL, Databend, BigQuery, Redshift, Snowflake, Amazon Athena, Firebolt, Azure Synapse Analytics

Infrastructure Required (external cluster needed):

  • Trino, Presto, Spark, LakeSail, Gluten+Velox (Linux/Docker), StarRocks, Doris, QuestDB, ClickHouse (server mode)

By Use Case

Local Development & Testing:

  • DuckDB (recommended), SQLite, DataFusion, Polars

Production Benchmarking:

  • Databricks, Snowflake, BigQuery, Redshift

Self-Hosted Analytics:

  • ClickHouse, StarRocks, Doris, QuestDB, PostgreSQL, Trino, Presto, Spark, LakeSail, Gluten+Velox

GPU Workloads:

  • CUDF (NVIDIA GPUs required)

Quick Start by Platform

Local Platforms (No Setup Required)

Terminal window
benchbox run --platform duckdb --benchmark tpch --scale 0.01
benchbox run --platform sqlite --benchmark tpch --scale 0.01

DuckDB is the default and is included in the base install. SQLite is also included in the base install.

Cloud Platforms (Credentials Required)

Terminal window
benchbox run --platform databricks --benchmark tpch --scale 1.0
benchbox run --platform bigquery --benchmark tpch --scale 1.0
benchbox run --platform snowflake --benchmark tpch --scale 1.0

Databricks requires DATABRICKS_TOKEN and DATABRICKS_HOST.

For BigQuery, Snowflake and Redshift, benchbox setup --platform <name> saves the credentials, and benchbox run reads those saved credentials rather than environment variables. If none are saved, run offers to start setup (and stops in non-interactive mode). Setup can fill its answers from environment variables when you accept auto-detection:

  • BigQuery setup needs both GOOGLE_APPLICATION_CREDENTIALS (a service account JSON key file) and BIGQUERY_PROJECT; BIGQUERY_DATASET, BIGQUERY_LOCATION and BIGQUERY_STORAGE_BUCKET are optional. Without them, setup prompts for the project ID and key file path.
  • Snowflake setup reads SNOWFLAKE_ACCOUNT, SNOWFLAKE_USERNAME, SNOWFLAKE_PASSWORD, SNOWFLAKE_WAREHOUSE and SNOWFLAKE_DATABASE.
  • Redshift setup reads REDSHIFT_HOST, REDSHIFT_USERNAME, REDSHIFT_PASSWORD and REDSHIFT_DATABASE.
Terminal window
benchbox setup --platform bigquery
benchbox run --platform bigquery --benchmark tpch --scale 1.0

Future Platforms